Compare commits

..
Author SHA1 Message Date
Kilian Schüttler c324d3815e Merge branch 'RED-8550-pre-backup' into 'main'
RED-8550: Faulty table recognition and text duplication leads to huge sections

See merge request fforesight/layout-parser!105
2024-02-28 16:13:56 +01:00
Maverick Studer 74f55a5cbf RED-8550: Faulty table recognition and text duplication leads to huge sections 2024-02-28 16:13:56 +01:00
Kilian Schüttler e7bf607663 Merge branch 'hotfix' into 'main'
hotfix: double viewerdoc writes in rare cases lead to some contentstreams not being written

See merge request fforesight/layout-parser!104
2024-02-26 12:32:14 +01:00
Kilian Schuettler f4d789311c hotfix: double viewerdoc writes in rare cases lead to some contentstreams not being written 2024-02-26 12:24:15 +01:00
Yannik Hampe 9817eae897 Merge branch 'RED-8481-fix' into 'main'
Red 8481 fix

See merge request fforesight/layout-parser!103
2024-02-23 14:08:57 +01:00
yhampe 477f6af886 RED-8481: Use visual layout parsing to detect signatures
added a new layer for visual parsing results

checkstyle
2024-02-23 14:02:53 +01:00
yhampe 2c171b6a9e RED-8481: Use visual layout parsing to detect signatures
added a new layer for visual parsing results

codestyle
2024-02-23 13:55:11 +01:00
yhampe 71477dabde RED-8481: Use visual layout parsing to detect signatures
added a new layer for visual parsing results

codestyle
2024-02-23 12:46:51 +01:00
yhampe a927cbd9dc RED-8481: Use visual layout parsing to detect signatures
added a new layer for visual parsing results

fixed tests
2024-02-23 12:38:05 +01:00
yhampe a1521877d7 RED-8481: Use visual layout parsing to detect signatures
added a new layer for visual parsing results

added a source label to image properties to enable rules
2024-02-23 12:20:11 +01:00
Maverick Studer f4b6386e1c Merge branch 'RED-8550-testing' into 'main'
RED-8550: Faulty table recognition and text duplication leads to huge sections

See merge request fforesight/layout-parser!102
2024-02-21 13:54:30 +01:00
Maverick Studer 1d64028158 RED-8550: Faulty table recognition and text duplication leads to huge sections 2024-02-21 13:54:30 +01:00
Yannik Hampe 0979a267d4 Merge branch 'RED-8481' into 'main'
RED-8481: Use visual layout parsing to detect signatures

See merge request fforesight/layout-parser!101
2024-02-15 13:07:06 +01:00
yhampe cc77d19500 RED-8481: Use visual layout parsing to detect signatures
addressed review comments
2024-02-15 13:01:30 +01:00
yhampe fa048b2fe0 RED-8481: Use visual layout parsing to detect signatures
addressed review comments
2024-02-15 12:19:26 +01:00
yhampe bdf1161c91 RED-8481: Use visual layout parsing to detect signatures
addressed review comments
2024-02-15 12:12:23 +01:00
yhampe b4a225144d RED-8481: Use visual layout parsing to detect signatures
working on failing tests
2024-02-15 10:16:07 +01:00
yhampe 903b1c1fd4 RED-8481: Use visual layout parsing to detect signatures
fixed failing tests because of null pointer
2024-02-15 09:27:07 +01:00
yhampe c3e7582ee3 RED-8481: Use visual layout parsing to detect signatures
fixed failing tests because of null pointer
2024-02-14 12:33:36 +01:00
yhampe cfc5db45cd RED-8481: Use visual layout parsing to detect signatures
fixed failing tests because of null pointer
2024-02-14 12:24:32 +01:00
yhampe fbd0196719 RED-8481: Use visual layout parsing to detect signatures
implemented visuallayoutparsingresult
2024-02-14 12:16:37 +01:00
Kilian Schüttler 3c9049dc8a Merge branch 'RED-8156' into 'main'
RED-8156: refactor ViewerDocumentService as a dependency for ocr-service

See merge request fforesight/layout-parser!99
2024-02-07 10:42:21 +01:00
Kilian Schuettler 015984891f RED-8156: refactor ViewerDocumentService as a dependency for ocr-service
* fix pmd
2024-02-06 17:17:26 +01:00
Kilian Schuettler 66fcb62833 RED-8156: refactor ViewerDocumentService as a dependency for ocr-service
* fix pmd
2024-02-06 17:09:21 +01:00
Kilian Schuettler 48824f56a8 RED-8156: refactor ViewerDocumentService as a dependency for ocr-service
* fix pmd
2024-02-06 17:06:53 +01:00
Kilian Schuettler 785628537f RED-8156: refactor ViewerDocumentService as a dependency for ocr-service
* various improvements to experimental parsing steps
* added embed fonts functionality to viewer doc
* fix checkstyle
2024-02-06 17:03:38 +01:00
Kilian Schuettler 23eb0c40a3 RED-8156: refactor ViewerDocumentService as a dependency for ocr-service
* various improvements to experimental parsing steps
* added embed fonts functionality to viewer doc
2024-02-06 16:59:51 +01:00
Dominique Eifländer 1b4aaf4454 Merge branch 'RED-8171' into 'main'
RED-8171: Traces do not stop at @Async

See merge request fforesight/layout-parser!98
2024-02-02 13:34:44 +01:00
Dominique Eifländer e4f3557b36 RED-8171: Traces do not stop at @Async 2024-02-02 13:22:57 +01:00
Timo Bejan 9be3c86297 Merge branch 'RED-8085' into 'main'
Red 8085

See merge request fforesight/layout-parser!96
2024-01-29 10:31:36 +01:00
Timo Bejan 88855de2da Red 8085 2024-01-29 10:31:36 +01:00
Dominique Eifländer 368a75e985 Merge branch 'RED-8106' into 'main'
RED-8106: Make documentdata serializable

See merge request fforesight/layout-parser!95
2023-12-22 13:33:02 +01:00
Dominique Eifländer 12344d57b2 RED-8106: Make documentdata serializable 2023-12-21 13:42:25 +01:00
97 changed files with 2759 additions and 1076 deletions
+5 -4
View File
@@ -9,12 +9,13 @@
</description>
<rule ref="category/java/errorprone.xml">
<exclude name="DataflowAnomalyAnalysis"/>
<exclude name="MissingSerialVersionUID"/>
<exclude name="NullAssignment"/>
<exclude name="AvoidLiteralsInIfCondition"/>
<exclude name="AvoidDuplicateLiterals"/>
<exclude name="AvoidFieldNameMatchingMethodName"/>
<exclude name="NullAssignment"/>
<exclude name="AssignmentInOperand"/>
<exclude name="BeanMembersShouldSerialize"/>
</rule>
</ruleset>
</ruleset>
+4 -4
View File
@@ -10,14 +10,14 @@
<rule ref="category/java/errorprone.xml">
<exclude name="DataflowAnomalyAnalysis"/>
<exclude name="MissingSerialVersionUID"/>
<exclude name="NullAssignment"/>
<exclude name="AvoidLiteralsInIfCondition"/>
<exclude name="AvoidDuplicateLiterals"/>
<exclude name="AvoidFieldNameMatchingMethodName"/>
<exclude name="AvoidFieldNameMatchingTypeName"/>
<exclude name="NullAssignment"/>
<exclude name="AssignmentInOperand"/>
<exclude name="TestClassWithoutTestCases"/>
<exclude name="BeanMembersShouldSerialize"/>
</rule>
</ruleset>
</ruleset>
@@ -1,6 +1,6 @@
plugins {
id("com.knecon.fforesight.java-conventions")
id("io.freefair.lombok") version "8.2.2"
id("io.freefair.lombok") version "8.4"
}
description = "layoutparser-service-internal-api"
@@ -1,5 +1,7 @@
package com.knecon.fforesight.service.layoutparser.internal.api.data.redaction;
import java.io.Serializable;
import io.swagger.v3.oas.annotations.media.Schema;
import lombok.AccessLevel;
import lombok.AllArgsConstructor;
@@ -12,7 +14,7 @@ import lombok.experimental.FieldDefaults;
@AllArgsConstructor
@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
@Schema(description = "Object containing the complete document layout parsing information. It is split into 4 categories, structure, text, positions and pages: " + "The document tree structure of SemanticNodes such as Section, Paragraph, Headline, etc. " + "The text, which is stored as separate blocks of data. " + "The text positions, which are also stored as separate blocks. The Blocks are equal to the text blocks in length and order. " + "The page information.")
public class DocumentData {
public class DocumentData implements Serializable {
@Schema(description = "Contains information about the document's pages.")
DocumentPage[] documentPages;
@@ -1,5 +1,7 @@
package com.knecon.fforesight.service.layoutparser.internal.api.data.redaction;
import java.io.Serializable;
import io.swagger.v3.oas.annotations.media.Schema;
import lombok.AccessLevel;
import lombok.AllArgsConstructor;
@@ -14,7 +16,7 @@ import lombok.experimental.FieldDefaults;
@AllArgsConstructor
@FieldDefaults(level = AccessLevel.PRIVATE)
@Schema(description = "Object containing information about the document's pages.")
public class DocumentPage {
public class DocumentPage implements Serializable {
@Schema(description = "The page number, starting with 1.")
int number;
@@ -1,5 +1,7 @@
package com.knecon.fforesight.service.layoutparser.internal.api.data.redaction;
import java.io.Serializable;
import io.swagger.v3.oas.annotations.media.Schema;
import lombok.AccessLevel;
import lombok.AllArgsConstructor;
@@ -14,7 +16,7 @@ import lombok.experimental.FieldDefaults;
@AllArgsConstructor
@FieldDefaults(level = AccessLevel.PRIVATE)
@Schema(description = "Object containing text positional information of a specific text block. A document is split into multiple text blocks, which are supposed to be read in order. Every text block can only occur on a single page.")
public class DocumentPositionData {
public class DocumentPositionData implements Serializable {
@Schema(description = "Identifier of the text block.")
Long id;
@@ -1,9 +1,11 @@
package com.knecon.fforesight.service.layoutparser.internal.api.data.redaction;
import java.awt.geom.Rectangle2D;
import java.io.Serializable;
import java.util.Arrays;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.stream.Stream;
import io.swagger.v3.oas.annotations.media.Schema;
@@ -20,14 +22,13 @@ import lombok.experimental.FieldDefaults;
@AllArgsConstructor
@FieldDefaults(level = AccessLevel.PRIVATE)
@Schema(description = "Object containing information about the parsed tree structure of the SemanticNodes, such as Section, Paragraph, Headline etc inside of the document.")
public class DocumentStructure {
public class DocumentStructure implements Serializable {
@Schema(description = "The root EntryData represents the Document.")
EntryData root;
@Schema(description = "Object containing the extra field names, a table has in its properties field.")
public static class TableProperties {
public static class TableProperties implements Serializable {
public static final String NUMBER_OF_ROWS = "numberOfRows";
public static final String NUMBER_OF_COLS = "numberOfCols";
@@ -35,7 +36,7 @@ public class DocumentStructure {
}
@Schema(description = "Object containing the extra field names, an Image has in its properties field.")
public static class ImageProperties {
public static class ImageProperties implements Serializable {
public static final String TRANSPARENT = "transparent";
public static final String IMAGE_TYPE = "imageType";
@@ -45,7 +46,7 @@ public class DocumentStructure {
}
@Schema(description = "Object containing the extra field names, a table cell has in its properties field.")
public static class TableCellProperties {
public static class TableCellProperties implements Serializable {
public static final String B_BOX = "bBox";
public static final String ROW = "row";
@@ -56,6 +57,7 @@ public class DocumentStructure {
public static final String RECTANGLE_DELIMITER = ";";
public static Rectangle2D parseRectangle2D(String bBox) {
List<Float> floats = Arrays.stream(bBox.split(RECTANGLE_DELIMITER)).map(Float::parseFloat).toList();
@@ -100,7 +102,7 @@ public class DocumentStructure {
@AllArgsConstructor
@FieldDefaults(level = AccessLevel.PRIVATE)
@Schema(description = "Object containing information of a SemanticNode and also structuring the layout with children.")
public static class EntryData {
public static class EntryData implements Serializable {
@Schema(description = "Type of the semantic node.", allowableValues = {"DOCUMENT", "SECTION", "PARAGRAPH", "HEADLINE", "TABLE", "TABLE_CELL", "HEADER", "FOOTER", "IMAGE"})
NodeType type;
@@ -114,6 +116,8 @@ public class DocumentStructure {
Map<String, String> properties;
@Schema(description = "All child Entries of this Entry.", example = "[1, 2, 3]")
List<EntryData> children;
@Schema(description = "Describes the origin of the semantic node", example = "[ALGORITHM]")
Set<LayoutEngine> engines;
@Override
@@ -1,5 +1,7 @@
package com.knecon.fforesight.service.layoutparser.internal.api.data.redaction;
import java.io.Serializable;
import io.swagger.v3.oas.annotations.media.Schema;
import lombok.AccessLevel;
import lombok.AllArgsConstructor;
@@ -14,7 +16,7 @@ import lombok.experimental.FieldDefaults;
@AllArgsConstructor
@FieldDefaults(level = AccessLevel.PRIVATE)
@Schema(description = "Object containing text information of a specific text block. A document is split into multiple text blocks, which are supposed to be read in order. Every text block can only occur on a single page.")
public class DocumentTextData {
public class DocumentTextData implements Serializable {
@Schema(description = "Identifier of the text block.")
Long id;
@@ -0,0 +1,6 @@
package com.knecon.fforesight.service.layoutparser.internal.api.data.redaction;
public enum LayoutEngine {
ALGORITHM,
AI
}
@@ -1,8 +1,9 @@
package com.knecon.fforesight.service.layoutparser.internal.api.data.redaction;
import java.io.Serializable;
import java.util.Locale;
public enum NodeType {
public enum NodeType implements Serializable {
DOCUMENT,
SECTION,
HEADLINE,
@@ -18,11 +18,15 @@ public record LayoutParsingRequest(
@Schema(description = "Path to the original PDF file.")//
@NonNull String originFileStorageId,//
@Schema(description = "Optional Path to the table extraction file.")//
Optional<String> tablesFileStorageId,//
@Schema(description = "Optional Path to the image classification file.")//
Optional<String> imagesFileStorageId,//
@Schema(description = "Optional Path to the the visual layout parsing service file") Optional<String> visualLayoutParsingFileId,//
@Schema(description = "Path where the Document Structure File will be stored.")//
@NonNull String structureFileStorageId,//
@Schema(description = "Path where the Research Data File will be stored.")//
@@ -1,6 +1,6 @@
plugins {
id("com.knecon.fforesight.java-conventions")
id("io.freefair.lombok") version "8.2.2"
id("io.freefair.lombok") version "8.4"
}
description = "layoutparser-service-processor"
@@ -10,12 +10,13 @@ val pdfBoxVersion = "3.0.0"
dependencies {
implementation(project(":layoutparser-service-internal-api"))
implementation(project(":viewer-doc-processor"))
implementation("com.iqser.red.service:persistence-service-shared-api-v1:2.144.0") {
exclude("org.springframework.boot", "spring-boot-starter-security")
exclude("org.springframework.boot", "spring-boot-starter-validation")
}
implementation("com.knecon.fforesight:tenant-commons:0.19.0")
implementation("com.knecon.fforesight:tenant-commons:0.21.0")
implementation("com.iqser.red.commons:storage-commons:2.45.0")
implementation("org.apache.pdfbox:pdfbox:${pdfBoxVersion}")
@@ -33,9 +33,11 @@ import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageB
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPositionSequence;
import com.knecon.fforesight.service.layoutparser.processor.python_api.adapter.CvTableParsingAdapter;
import com.knecon.fforesight.service.layoutparser.processor.python_api.adapter.ImageServiceResponseAdapter;
import com.knecon.fforesight.service.layoutparser.processor.python_api.adapter.VisualLayoutParsingAdapter;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.image.ImageServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableCells;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.services.BodyTextFrameService;
import com.knecon.fforesight.service.layoutparser.processor.services.RulingCleaningService;
import com.knecon.fforesight.service.layoutparser.processor.services.SectionsBuilderService;
@@ -51,7 +53,7 @@ import com.knecon.fforesight.service.layoutparser.processor.services.factory.Doc
import com.knecon.fforesight.service.layoutparser.processor.services.mapper.DocumentDataMapper;
import com.knecon.fforesight.service.layoutparser.processor.services.mapper.TaasDocumentDataMapper;
import com.knecon.fforesight.service.layoutparser.processor.services.parsing.PDFLinesTextStripper;
import com.knecon.fforesight.service.layoutparser.processor.services.visualization.ViewerDocumentService;
import com.knecon.fforesight.service.layoutparser.processor.services.visualization.LayoutGridService;
import com.knecon.fforesight.service.layoutparser.processor.utils.MarkedContentUtils;
import io.micrometer.observation.Observation;
@@ -84,8 +86,9 @@ public class LayoutParsingPipeline {
TaasBlockificationService taasBlockificationService;
DocuMineBlockificationService docuMineBlockificationService;
RedactManagerBlockificationService redactManagerBlockificationService;
ViewerDocumentService viewerDocumentService;
LayoutGridService layoutGridService;
ObservationRegistry observationRegistry;
VisualLayoutParsingAdapter visualLayoutParsingAdapter;
public LayoutParsingFinishedEvent parseLayoutAndSaveFilesToStorage(LayoutParsingRequest layoutParsingRequest) throws IOException {
@@ -94,67 +97,86 @@ public class LayoutParsingPipeline {
log.info("Starting layout parsing for {}", layoutParsingRequest.identifier());
File originFile = layoutParsingStorageService.getOriginFile(layoutParsingRequest.originFileStorageId());
File viewerDocumentFile = File.createTempFile("viewer_document", ".pdf");
File viewerDocumentFile = layoutParsingStorageService.getViewerDocFile(layoutParsingRequest.viewerDocumentStorageId())
.orElse(originFile);
VisualLayoutParsingResponse visualLayoutParsingResponse = new VisualLayoutParsingResponse();
if (layoutParsingRequest.visualLayoutParsingFileId()
.isPresent()) {
visualLayoutParsingResponse = layoutParsingStorageService.getVisualLayoutParsingFile(layoutParsingRequest.visualLayoutParsingFileId()
.get());
}
ImageServiceResponse imageServiceResponse = new ImageServiceResponse();
if (layoutParsingRequest.imagesFileStorageId().isPresent()) {
imageServiceResponse = layoutParsingStorageService.getImagesFile(layoutParsingRequest.imagesFileStorageId().get());
if (layoutParsingRequest.imagesFileStorageId()
.isPresent()) {
imageServiceResponse = layoutParsingStorageService.getImagesFile(layoutParsingRequest.imagesFileStorageId()
.get());
}
TableServiceResponse tableServiceResponse = new TableServiceResponse();
if (layoutParsingRequest.tablesFileStorageId().isPresent()) {
tableServiceResponse = layoutParsingStorageService.getTablesFile(layoutParsingRequest.tablesFileStorageId().get());
if (layoutParsingRequest.tablesFileStorageId()
.isPresent()) {
tableServiceResponse = layoutParsingStorageService.getTablesFile(layoutParsingRequest.tablesFileStorageId()
.get());
}
ClassificationDocument classificationDocument = parseLayout(layoutParsingRequest.layoutParsingType(),
originFile,
imageServiceResponse,
tableServiceResponse,
layoutParsingRequest.identifier().toString());
originFile,
imageServiceResponse,
tableServiceResponse,
visualLayoutParsingResponse,
layoutParsingRequest.identifier().toString());
log.info("Building document graph for {}", layoutParsingRequest.identifier());
Document documentGraph = observeBuildDocumentGraph(classificationDocument);
log.info("Creating viewer document for {}", layoutParsingRequest.identifier());
layoutGridService.addLayoutGrid(viewerDocumentFile, documentGraph, viewerDocumentFile, false);
log.info("Storing resulting files for {}", layoutParsingRequest.identifier());
layoutParsingStorageService.storeDocumentData(layoutParsingRequest, DocumentDataMapper.toDocumentData(documentGraph));
layoutParsingStorageService.storeSimplifiedText(layoutParsingRequest, simplifiedSectionTextService.toSimplifiedText(documentGraph));
log.info("Creating viewer document for {}", layoutParsingRequest.identifier());
viewerDocumentService.createViewerDocument(originFile, documentGraph, viewerDocumentFile, false);
layoutParsingStorageService.storeViewerDocument(layoutParsingRequest, viewerDocumentFile);
if (layoutParsingRequest.layoutParsingType().equals(LayoutParsingType.TAAS)) {
log.info("Building research document data for {}", layoutParsingRequest.identifier());
var researchDocumentData = TaasDocumentDataMapper.fromDocument(documentGraph);
layoutParsingStorageService.storeResearchDocumentData(layoutParsingRequest, researchDocumentData);
}
if (!viewerDocumentFile.equals(originFile)) {
viewerDocumentFile.delete();
}
originFile.delete();
viewerDocumentFile.delete();
return LayoutParsingFinishedEvent.builder()
.identifier(layoutParsingRequest.identifier())
.numberOfPages(documentGraph.getNumberOfPages())
.duration(System.currentTimeMillis() - start)
.message(format("""
Layout parsing has finished in %.02f s.
identifiers: %s
%s
Files have been saved with Ids:
Structure: %s
Text: %s
Positions: %s
PageData: %s
Simplified Text: %s
Viewer Doc: %s""",
((float) (System.currentTimeMillis() - start)) / 1000,
layoutParsingRequest.identifier(),
buildSemanticNodeCountMessage(documentGraph.getNumberOfPages(), documentGraph.buildSemanticNodeCounts()),
layoutParsingRequest.structureFileStorageId(),
layoutParsingRequest.textBlockFileStorageId(),
layoutParsingRequest.positionBlockFileStorageId(),
layoutParsingRequest.pageFileStorageId(),
layoutParsingRequest.simplifiedTextStorageId(),
layoutParsingRequest.viewerDocumentStorageId()))
Layout parsing has finished in %.02f s.
identifiers: %s
%s
Files have been saved with Ids:
Structure: %s
Text: %s
Positions: %s
PageData: %s
Simplified Text: %s
Viewer Doc: %s""",
((float) (System.currentTimeMillis() - start)) / 1000,
layoutParsingRequest.identifier(),
buildSemanticNodeCountMessage(documentGraph.getNumberOfPages(), documentGraph.buildSemanticNodeCounts()),
layoutParsingRequest.structureFileStorageId(),
layoutParsingRequest.textBlockFileStorageId(),
layoutParsingRequest.positionBlockFileStorageId(),
layoutParsingRequest.pageFileStorageId(),
layoutParsingRequest.simplifiedTextStorageId(),
layoutParsingRequest.viewerDocumentStorageId()))
.build();
}
@@ -164,9 +186,9 @@ public class LayoutParsingPipeline {
AtomicReference<Document> documentReference = new AtomicReference<>();
Observation.createNotStarted("LayoutParsingPipeline", observationRegistry).contextualName("build-document-graph").observe(() -> {
documentReference.set(DocumentGraphFactory.buildDocumentGraph(classificationDocument));
});
Observation.createNotStarted("LayoutParsingPipeline", observationRegistry)
.contextualName("build-document-graph")
.observe(() -> documentReference.set(DocumentGraphFactory.buildDocumentGraph(classificationDocument)));
return documentReference.get();
}
@@ -175,14 +197,14 @@ public class LayoutParsingPipeline {
private String buildSemanticNodeCountMessage(int numberOfPages, Map<NodeType, Long> semanticNodeCounts) {
return String.format("%d pages with %d sections, %d headlines, %d paragraphs, %d tables with %d cells, %d headers, and %d footers parsed",
numberOfPages,
semanticNodeCounts.get(NodeType.SECTION) == null ? 0 : semanticNodeCounts.get(NodeType.SECTION),
semanticNodeCounts.get(NodeType.HEADLINE) == null ? 0 : semanticNodeCounts.get(NodeType.HEADLINE),
semanticNodeCounts.get(NodeType.PARAGRAPH) == null ? 0 : semanticNodeCounts.get(NodeType.PARAGRAPH),
semanticNodeCounts.get(NodeType.TABLE) == null ? 0 : semanticNodeCounts.get(NodeType.TABLE),
semanticNodeCounts.get(NodeType.TABLE_CELL) == null ? 0 : semanticNodeCounts.get(NodeType.TABLE_CELL),
semanticNodeCounts.get(NodeType.HEADER) == null ? 0 : semanticNodeCounts.get(NodeType.HEADER),
semanticNodeCounts.get(NodeType.FOOTER) == null ? 0 : semanticNodeCounts.get(NodeType.FOOTER));
numberOfPages,
semanticNodeCounts.get(NodeType.SECTION) == null ? 0 : semanticNodeCounts.get(NodeType.SECTION),
semanticNodeCounts.get(NodeType.HEADLINE) == null ? 0 : semanticNodeCounts.get(NodeType.HEADLINE),
semanticNodeCounts.get(NodeType.PARAGRAPH) == null ? 0 : semanticNodeCounts.get(NodeType.PARAGRAPH),
semanticNodeCounts.get(NodeType.TABLE) == null ? 0 : semanticNodeCounts.get(NodeType.TABLE),
semanticNodeCounts.get(NodeType.TABLE_CELL) == null ? 0 : semanticNodeCounts.get(NodeType.TABLE_CELL),
semanticNodeCounts.get(NodeType.HEADER) == null ? 0 : semanticNodeCounts.get(NodeType.HEADER),
semanticNodeCounts.get(NodeType.FOOTER) == null ? 0 : semanticNodeCounts.get(NodeType.FOOTER));
}
@@ -192,13 +214,14 @@ public class LayoutParsingPipeline {
File originFile,
ImageServiceResponse imageServiceResponse,
TableServiceResponse tableServiceResponse,
VisualLayoutParsingResponse visualLayoutParsingResponse,
String identifier) {
PDDocument originDocument = openDocument(originFile);
addNumberOfPagesToTrace(originDocument.getNumberOfPages(), Files.size(originFile.toPath()));
Map<Integer, List<TableCells>> pdfTableCells = cvTableParsingAdapter.buildCvParsedTablesPerPage(tableServiceResponse);
Map<Integer, List<ClassifiedImage>> pdfImages = imageServiceResponseAdapter.buildClassifiedImagesPerPage(imageServiceResponse);
Map<Integer, List<ClassifiedImage>> signatures = visualLayoutParsingAdapter.buildExtractedSignaturesPerPage(visualLayoutParsingResponse);
ClassificationDocument classificationDocument = new ClassificationDocument();
List<ClassificationPage> classificationPages = new ArrayList<>();
@@ -258,6 +281,14 @@ public class LayoutParsingPipeline {
imageServiceResponseAdapter.findOcr(classificationPage);
}
if (signatures.containsKey(pageNumber)) {
if (classificationPage.getImages() == null || classificationPage.getImages().size() == 0) {
classificationPage.setImages(signatures.get(pageNumber));
} else {
classificationPage.getImages().addAll(signatures.get(pageNumber));
}
}
tableExtractionService.extractTables(cleanRulings, classificationPage);
buildPageStatistics(classificationPage);
@@ -313,9 +344,7 @@ public class LayoutParsingPipeline {
private void increaseDocumentStatistics(ClassificationPage classificationPage, ClassificationDocument document) {
if (!classificationPage.isLandscape()) {
document.getFontSizeCounter().addAll(classificationPage.getFontSizeCounter().getCountPerValue());
}
document.getFontSizeCounter().addAll(classificationPage.getFontSizeCounter().getCountPerValue());
document.getFontCounter().addAll(classificationPage.getFontCounter().getCountPerValue());
document.getTextHeightCounter().addAll(classificationPage.getTextHeightCounter().getCountPerValue());
document.getFontStyleCounter().addAll(classificationPage.getFontStyleCounter().getCountPerValue());
@@ -1,10 +1,24 @@
package com.knecon.fforesight.service.layoutparser.processor;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.ComponentScan;
import org.springframework.context.annotation.Configuration;
import com.knecon.fforesight.service.viewerdoc.service.ViewerDocumentService;
import io.micrometer.observation.ObservationRegistry;
@Configuration
@ComponentScan
public class LayoutParsingServiceProcessorConfiguration {
@Bean
@Autowired
public ViewerDocumentService viewerDocumentService(ObservationRegistry registry) {
return new ViewerDocumentService(registry);
}
}
@@ -8,9 +8,8 @@ import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.nio.file.StandardOpenOption;
import java.util.Optional;
import org.apache.pdfbox.Loader;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.springframework.stereotype.Service;
import com.fasterxml.jackson.databind.ObjectMapper;
@@ -21,6 +20,7 @@ import com.knecon.fforesight.service.layoutparser.internal.api.data.taas.Researc
import com.knecon.fforesight.service.layoutparser.internal.api.queue.LayoutParsingRequest;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.image.ImageServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.tenantcommons.TenantContext;
import io.micrometer.observation.annotation.Observed;
@@ -36,13 +36,6 @@ public class LayoutParsingStorageService {
private final StorageService storageService;
private final ObjectMapper objectMapper;
public PDDocument getOriginDocument(String storageId) throws IOException {
return Loader.loadPDF(getOriginFile(storageId));
}
@Observed(name = "LayoutParsingStorageService", contextualName = "get-origin-file")
public File getOriginFile(String storageId) throws IOException {
@@ -52,6 +45,18 @@ public class LayoutParsingStorageService {
}
@Observed(name = "LayoutParsingStorageService", contextualName = "get-viewer-doc-file")
public Optional<File> getViewerDocFile(String storageId) throws IOException {
if (!storageService.objectExists(TenantContext.getTenantId(), storageId)) {
return Optional.empty();
}
File tempFile = createTempFile("viewerDocument", ".pdf");
storageService.downloadTo(TenantContext.getTenantId(), storageId, tempFile);
return Optional.of(tempFile);
}
public ImageServiceResponse getImagesFile(String storageId) throws IOException {
try (InputStream inputStream = getObject(storageId)) {
@@ -73,6 +78,14 @@ public class LayoutParsingStorageService {
}
}
public VisualLayoutParsingResponse getVisualLayoutParsingFile(String storageId) throws IOException {
try (InputStream inputStream = getObject(storageId)) {
VisualLayoutParsingResponse visualLayoutParsingResponse = objectMapper.readValue(inputStream, VisualLayoutParsingResponse.class);
return visualLayoutParsingResponse;
}
}
@Observed(name = "LayoutParsingStorageService", contextualName = "store-document-data")
public void storeDocumentData(LayoutParsingRequest layoutParsingRequest, DocumentData documentData) {
@@ -137,7 +150,6 @@ public class LayoutParsingStorageService {
public void storeViewerDocument(LayoutParsingRequest layoutParsingRequest, File out) {
try (var in = new FileInputStream(out)) {
storageService.storeObject(TenantContext.getTenantId(), layoutParsingRequest.viewerDocumentStorageId(), in);
}
}
@@ -6,12 +6,14 @@ import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageB
import lombok.AllArgsConstructor;
import lombok.Data;
import lombok.EqualsAndHashCode;
import lombok.NoArgsConstructor;
@Data
@AllArgsConstructor
@NoArgsConstructor
public abstract class AbstractPageBlock {
@EqualsAndHashCode(callSuper = true)
public abstract class AbstractPageBlock extends Rectangle {
@JsonIgnore
protected float minX;
@@ -12,6 +12,7 @@ import lombok.Setter;
@Setter
@EqualsAndHashCode
@SuppressWarnings("PMD.AvoidFieldNameMatchingMethodName")
public class Boundary implements Comparable<Boundary> {
private int start;
@@ -10,6 +10,7 @@ import java.util.Set;
import java.util.stream.Collectors;
import java.util.stream.Stream;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -30,6 +31,9 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Document implements GenericSemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
Set<Page> pages;
DocumentTree documentTree;
Integer numberOfPages;
@@ -56,13 +60,15 @@ public class Document implements GenericSemanticNode {
public List<Section> getMainSections() {
return streamChildrenOfType(NodeType.SECTION).map(node -> (Section) node).collect(Collectors.toList());
return streamChildrenOfType(NodeType.SECTION).map(node -> (Section) node)
.collect(Collectors.toList());
}
public Stream<TextBlock> streamTerminalTextBlocksInOrder() {
return streamAllNodes().filter(SemanticNode::isLeaf).map(SemanticNode::getLeafTextBlock);
return streamAllNodes().filter(SemanticNode::isLeaf)
.map(SemanticNode::getLeafTextBlock);
}
@@ -83,13 +89,16 @@ public class Document implements GenericSemanticNode {
@Override
public Headline getHeadline() {
return streamAllSubNodesOfType(NodeType.HEADLINE).map(node -> (Headline) node).findFirst().orElse(Headline.builder().build());
return streamAllSubNodesOfType(NodeType.HEADLINE).map(node -> (Headline) node)
.findFirst()
.orElse(Headline.builder().build());
}
private Stream<SemanticNode> streamAllNodes() {
return documentTree.allEntriesInOrder().map(DocumentTree.Entry::getNode);
return documentTree.allEntriesInOrder()
.map(DocumentTree.Entry::getNode);
}
@@ -6,6 +6,7 @@ import java.util.List;
import java.util.Map;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -26,6 +27,9 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Footer implements GenericSemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
List<Integer> treeId;
TextBlock leafTextBlock;
@@ -6,6 +6,7 @@ import java.util.List;
import java.util.Map;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -26,6 +27,8 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Header implements GenericSemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
List<Integer> treeId;
TextBlock leafTextBlock;
@@ -6,6 +6,7 @@ import java.util.List;
import java.util.Map;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -26,6 +27,8 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Headline implements GenericSemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
List<Integer> treeId;
TextBlock leafTextBlock;
@@ -8,6 +8,7 @@ import java.util.List;
import java.util.Map;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -29,6 +30,9 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Image implements GenericSemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
List<Integer> treeId;
String id;
@@ -66,7 +70,9 @@ public class Image implements GenericSemanticNode {
@Override
public TextBlock getTextBlock() {
return streamAllSubNodes().filter(SemanticNode::isLeaf).map(SemanticNode::getLeafTextBlock).collect(new TextBlockCollector());
return streamAllSubNodes().filter(SemanticNode::isLeaf)
.map(SemanticNode::getLeafTextBlock)
.collect(new TextBlockCollector());
}
@@ -6,6 +6,8 @@ public enum ImageType {
LOGO,
FORMULA,
SIGNATURE,
SIGNATURE_VISUAL,
OTHER,
OCR;
@@ -6,6 +6,7 @@ import java.util.List;
import java.util.Map;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -24,6 +25,9 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Paragraph implements GenericSemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
List<Integer> treeId;
TextBlock leafTextBlock;
@@ -6,6 +6,7 @@ import java.util.List;
import java.util.Map;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -27,6 +28,8 @@ import lombok.extern.slf4j.Slf4j;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Section implements GenericSemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
List<Integer> treeId;
TextBlock textBlock;
@@ -50,7 +53,8 @@ public class Section implements GenericSemanticNode {
public boolean hasTables() {
return streamAllSubNodesOfType(NodeType.TABLE).findAny().isPresent();
return streamAllSubNodesOfType(NodeType.TABLE).findAny()
.isPresent();
}
@@ -58,7 +62,9 @@ public class Section implements GenericSemanticNode {
public TextBlock getTextBlock() {
if (textBlock == null) {
textBlock = streamAllSubNodes().filter(SemanticNode::isLeaf).map(SemanticNode::getLeafTextBlock).collect(new TextBlockCollector());
textBlock = streamAllSubNodes().filter(SemanticNode::isLeaf)
.map(SemanticNode::getLeafTextBlock)
.collect(new TextBlockCollector());
}
return textBlock;
}
@@ -12,6 +12,7 @@ import java.util.Set;
import java.util.stream.Collectors;
import java.util.stream.Stream;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.Boundary;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
@@ -335,6 +336,23 @@ public interface SemanticNode {
}
/**
* returns the set of layoutengines.
*
* @return set of layoutengines.
*/
Set<LayoutEngine> getEngines();
/**
* adds a layoutengine to the set.
*/
default void addEngine(LayoutEngine engine) {
getEngines().add(engine);
}
/**
* Streams all children located directly underneath this node in the DocumentTree.
*
@@ -426,6 +444,7 @@ public interface SemanticNode {
/**
* TODO: this produces unwanted results for sections spanning multiple columns.
* Computes the Union of the bounding boxes of all children recursively.
*
* @return The union of the BoundingBoxes of all children
*/
private Map<Page, Rectangle2D> getBBoxFromChildren() {
@@ -12,6 +12,7 @@ import java.util.Set;
import java.util.stream.IntStream;
import java.util.stream.Stream;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -31,6 +32,8 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Table implements SemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
List<Integer> treeId;
DocumentTree documentTree;
@@ -45,6 +48,7 @@ public class Table implements SemanticNode {
@EqualsAndHashCode.Exclude
Map<Page, Rectangle2D> bBoxCache;
/**
* Streams all entities in this table, that appear in a row, which contains any of the provided strings.
*
@@ -53,8 +57,7 @@ public class Table implements SemanticNode {
*/
public Stream<RedactionEntity> streamEntitiesWhereRowContainsStringsIgnoreCase(List<String> strings) {
return IntStream.range(0, numberOfRows)
.boxed()
return IntStream.range(0, numberOfRows).boxed()
.filter(row -> rowContainsStringsIgnoreCase(row, strings))
.flatMap(this::streamRow)
.map(TableCell::getEntities)
@@ -71,8 +74,11 @@ public class Table implements SemanticNode {
*/
public boolean rowContainsStringsIgnoreCase(Integer row, List<String> strings) {
String rowText = streamRow(row).map(TableCell::getTextBlock).collect(new TextBlockCollector()).getSearchText().toLowerCase(Locale.ROOT);
return strings.stream().map(String::toLowerCase).allMatch(rowText::contains);
String rowText = streamRow(row).map(TableCell::getTextBlock)
.collect(new TextBlockCollector()).getSearchText().toLowerCase(Locale.ROOT);
return strings.stream()
.map(String::toLowerCase)
.allMatch(rowText::contains);
}
@@ -85,9 +91,13 @@ public class Table implements SemanticNode {
*/
public Stream<RedactionEntity> streamEntitiesWhereRowHasHeaderAndValue(String header, String value) {
List<Integer> vertebrateStudyCols = streamHeaders().filter(headerNode -> headerNode.containsString(header)).map(TableCell::getCol).toList();
List<Integer> vertebrateStudyCols = streamHeaders().filter(headerNode -> headerNode.containsString(header))
.map(TableCell::getCol)
.toList();
return streamTableCells().filter(tableCellNode -> vertebrateStudyCols.stream()
.anyMatch(vertebrateStudyCol -> getCell(tableCellNode.getRow(), vertebrateStudyCol).containsString(value))).map(TableCell::getEntities).flatMap(Collection::stream);
.anyMatch(vertebrateStudyCol -> getCell(tableCellNode.getRow(), vertebrateStudyCol).containsString(value)))
.map(TableCell::getEntities)
.flatMap(Collection::stream);
}
@@ -100,9 +110,13 @@ public class Table implements SemanticNode {
*/
public Stream<RedactionEntity> streamEntitiesWhereRowHasHeaderAndAnyValue(String header, List<String> values) {
List<Integer> colsWithHeader = streamHeaders().filter(headerNode -> headerNode.containsString(header)).map(TableCell::getCol).toList();
List<Integer> colsWithHeader = streamHeaders().filter(headerNode -> headerNode.containsString(header))
.map(TableCell::getCol)
.toList();
return streamTableCells().filter(tableCellNode -> colsWithHeader.stream()
.anyMatch(colWithHeader -> getCell(tableCellNode.getRow(), colWithHeader).containsAnyString(values))).map(TableCell::getEntities).flatMap(Collection::stream);
.anyMatch(colWithHeader -> getCell(tableCellNode.getRow(), colWithHeader).containsAnyString(values)))
.map(TableCell::getEntities)
.flatMap(Collection::stream);
}
@@ -114,12 +128,15 @@ public class Table implements SemanticNode {
*/
public Stream<RedactionEntity> streamEntitiesWhereRowContainsEntitiesOfType(List<String> types) {
List<Integer> rowsWithEntityOfType = IntStream.range(0, numberOfRows)
.boxed()
.filter(rowNumber -> streamEntityTypesInRow(rowNumber).anyMatch(existingType -> types.stream().anyMatch(typeToCheck -> typeToCheck.equals(existingType))))
List<Integer> rowsWithEntityOfType = IntStream.range(0, numberOfRows).boxed()
.filter(rowNumber -> streamEntityTypesInRow(rowNumber).anyMatch(existingType -> types.stream()
.anyMatch(typeToCheck -> typeToCheck.equals(existingType))))
.toList();
return rowsWithEntityOfType.stream().flatMap(this::streamRow).map(TableCell::getEntities).flatMap(Collection::stream);
return rowsWithEntityOfType.stream()
.flatMap(this::streamRow)
.map(TableCell::getEntities)
.flatMap(Collection::stream);
}
@@ -131,18 +148,24 @@ public class Table implements SemanticNode {
*/
public Stream<RedactionEntity> streamEntitiesWhereRowContainsNoEntitiesOfType(List<String> types) {
List<Integer> rowsWithNoEntityOfType = IntStream.range(0, numberOfRows)
.boxed()
.filter(rowNumber -> streamEntityTypesInRow(rowNumber).noneMatch(existingType -> types.stream().anyMatch(typeToCheck -> typeToCheck.equals(existingType))))
List<Integer> rowsWithNoEntityOfType = IntStream.range(0, numberOfRows).boxed()
.filter(rowNumber -> streamEntityTypesInRow(rowNumber).noneMatch(existingType -> types.stream()
.anyMatch(typeToCheck -> typeToCheck.equals(existingType))))
.toList();
return rowsWithNoEntityOfType.stream().flatMap(this::streamRow).map(TableCell::getEntities).flatMap(Collection::stream);
return rowsWithNoEntityOfType.stream()
.flatMap(this::streamRow)
.map(TableCell::getEntities)
.flatMap(Collection::stream);
}
private Stream<String> streamEntityTypesInRow(Integer rowNumber) {
return streamRow(rowNumber).map(TableCell::getEntities).flatMap(Collection::stream).map(RedactionEntity::getType).distinct();
return streamRow(rowNumber).map(TableCell::getEntities)
.flatMap(Collection::stream)
.map(RedactionEntity::getType)
.distinct();
}
@@ -159,7 +182,8 @@ public class Table implements SemanticNode {
throw new IllegalArgumentException(format("row %d, col %d is out of bounds for number of rows of %d and number of cols %d", row, col, numberOfRows, numberOfCols));
}
int idx = row * numberOfCols + col;
return (TableCell) documentTree.getEntryById(treeId).getChildren().get(idx).getNode();
return (TableCell) documentTree.getEntryById(treeId).getChildren()
.get(idx).getNode();
}
@@ -196,7 +220,8 @@ public class Table implements SemanticNode {
*/
public Stream<TableCell> streamCol(int col) {
return IntStream.range(0, numberOfRows).boxed().map(row -> getCell(row, col));
return IntStream.range(0, numberOfRows).boxed()
.map(row -> getCell(row, col));
}
@@ -208,9 +233,11 @@ public class Table implements SemanticNode {
*/
public Stream<TableCell> streamRow(int row) {
return IntStream.range(0, numberOfCols).boxed().map(col -> getCell(row, col));
return IntStream.range(0, numberOfCols).boxed()
.map(col -> getCell(row, col));
}
/**
* Streams all TableCells row-wise and filters them with header == true.
*
@@ -231,7 +258,8 @@ public class Table implements SemanticNode {
*/
public Stream<TableCell> streamHeadersForCell(int row, int col) {
return Stream.concat(streamRow(row), streamCol(col)).filter(TableCell::isHeader);
return Stream.concat(streamRow(row), streamCol(col))
.filter(TableCell::isHeader);
}
@@ -304,7 +332,9 @@ public class Table implements SemanticNode {
public TextBlock getTextBlock() {
if (textBlock == null) {
textBlock = streamAllSubNodes().filter(SemanticNode::isLeaf).map(SemanticNode::getLeafTextBlock).collect(new TextBlockCollector());
textBlock = streamAllSubNodes().filter(SemanticNode::isLeaf)
.map(SemanticNode::getLeafTextBlock)
.collect(new TextBlockCollector());
}
return textBlock;
}
@@ -315,6 +345,8 @@ public class Table implements SemanticNode {
return treeId.toString() + ": " + NodeType.TABLE + ": #cols: " + numberOfCols + ", #rows: " + numberOfRows + ", " + this.getTextBlock().buildSummary();
}
@Override
public Map<Page, Rectangle2D> getBBox() {
@@ -323,4 +355,5 @@ public class Table implements SemanticNode {
}
return bBoxCache;
}
}
@@ -7,6 +7,7 @@ import java.util.List;
import java.util.Map;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.RedactionEntity;
@@ -26,6 +27,8 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(level = AccessLevel.PRIVATE)
public class TableCell implements GenericSemanticNode {
@Builder.Default
Set<LayoutEngine> engines = new HashSet<>(Set.of(LayoutEngine.ALGORITHM));
List<Integer> treeId;
int row;
int col;
@@ -84,14 +87,16 @@ public class TableCell implements GenericSemanticNode {
private TextBlock buildTextBlock() {
return streamAllSubNodes().filter(SemanticNode::isLeaf).map(SemanticNode::getLeafTextBlock).collect(new TextBlockCollector());
return streamAllSubNodes().filter(SemanticNode::isLeaf)
.map(SemanticNode::getLeafTextBlock)
.collect(new TextBlockCollector());
}
@Override
public String toString() {
return treeId + ": " + NodeType.TABLE_CELL + ": " + this.buildTextBlock().buildSummary();
return treeId + ": " + NodeType.TABLE_CELL + ": " + this.getTextBlock().buildSummary();
}
}
@@ -4,18 +4,21 @@ import java.awt.geom.Rectangle2D;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.ImageType;
import lombok.AllArgsConstructor;
import lombok.Data;
import lombok.NonNull;
import lombok.RequiredArgsConstructor;
@Data
@RequiredArgsConstructor
@AllArgsConstructor
public class ClassifiedImage {
@NonNull
private Rectangle2D position;
@NonNull
private ImageType imageType;
private boolean sourceByAi;
private boolean isAppendedToSection;
@NonNull
private boolean hasTransparency;
@@ -1,6 +1,7 @@
package com.knecon.fforesight.service.layoutparser.processor.model.table;
import java.awt.geom.Point2D;
import java.awt.geom.Rectangle2D;
import java.util.ArrayList;
import java.util.Iterator;
import java.util.List;
@@ -36,6 +37,12 @@ public class Cell extends Rectangle {
}
public Cell(Rectangle2D r) {
super((float) r.getY(), (float) r.getX(), (float) r.getWidth(), (float) r.getHeight());
}
public void addTextBlock(TextPageBlock textBlock) {
textBlocks.add(textBlock);
@@ -20,7 +20,8 @@ import lombok.extern.slf4j.Slf4j;
@SuppressWarnings("all")
public class Ruling extends Line2D.Float {
private static int PERPENDICULAR_PIXEL_EXPAND_AMOUNT = 2;
public static final int PERPENDICULAR_UNIT_EXPAND_AMOUNT = 2;
public static final int COLINEAR_OR_PARALLEL_UNIT_EXPAND_AMOUNT = 2;
public Ruling(Point2D p1, Point2D p2) {
@@ -110,8 +111,8 @@ public class Ruling extends Line2D.Float {
});
for (Ruling h : horizontals) {
sos.add(new SortObject(SOType.HLEFT, h.getLeft() - PERPENDICULAR_PIXEL_EXPAND_AMOUNT, h));
sos.add(new SortObject(SOType.HRIGHT, h.getRight() + PERPENDICULAR_PIXEL_EXPAND_AMOUNT, h));
sos.add(new SortObject(SOType.HLEFT, h.getLeft() - PERPENDICULAR_UNIT_EXPAND_AMOUNT, h));
sos.add(new SortObject(SOType.HRIGHT, h.getRight() + PERPENDICULAR_UNIT_EXPAND_AMOUNT, h));
}
for (Ruling v : verticals) {
@@ -151,7 +152,7 @@ public class Ruling extends Line2D.Float {
if (i == null) {
continue;
}
rv.put(i, new Ruling[]{h.getKey().expand(PERPENDICULAR_PIXEL_EXPAND_AMOUNT), so.ruling.expand(PERPENDICULAR_PIXEL_EXPAND_AMOUNT)});
rv.put(i, new Ruling[]{h.getKey().expand(PERPENDICULAR_UNIT_EXPAND_AMOUNT), so.ruling.expand(PERPENDICULAR_UNIT_EXPAND_AMOUNT)});
} catch (UnsupportedOperationException e) {
log.info("Some line are oblique, ignoring...");
continue;
@@ -267,7 +268,7 @@ public class Ruling extends Line2D.Float {
}
public boolean nearlyIntersects(Ruling another, int colinearOrParallelExpandAmount) {
public boolean nearlyIntersects(Ruling another) {
if (this.intersectsLine(another)) {
return true;
@@ -276,9 +277,9 @@ public class Ruling extends Line2D.Float {
boolean rv = false;
if (this.perpendicularTo(another)) {
rv = this.expand(PERPENDICULAR_PIXEL_EXPAND_AMOUNT).intersectsLine(another);
rv = this.expand(PERPENDICULAR_UNIT_EXPAND_AMOUNT).intersectsLine(another);
} else {
rv = this.expand(colinearOrParallelExpandAmount).intersectsLine(another.expand(colinearOrParallelExpandAmount));
rv = this.expand(COLINEAR_OR_PARALLEL_UNIT_EXPAND_AMOUNT).intersectsLine(another.expand(COLINEAR_OR_PARALLEL_UNIT_EXPAND_AMOUNT));
}
return rv;
@@ -319,8 +320,8 @@ public class Ruling extends Line2D.Float {
public Point2D intersectionPoint(Ruling other) {
Ruling this_l = this.expand(PERPENDICULAR_PIXEL_EXPAND_AMOUNT);
Ruling other_l = other.expand(PERPENDICULAR_PIXEL_EXPAND_AMOUNT);
Ruling this_l = this.expand(PERPENDICULAR_UNIT_EXPAND_AMOUNT);
Ruling other_l = other.expand(PERPENDICULAR_UNIT_EXPAND_AMOUNT);
Ruling horizontal, vertical;
if (!this_l.intersectsLine(other_l)) {
@@ -1,18 +1,18 @@
package com.knecon.fforesight.service.layoutparser.processor.model.table;
import java.awt.geom.Point2D;
import java.awt.geom.Rectangle2D;
import java.util.ArrayList;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashSet;
import java.util.List;
import java.util.Set;
import java.util.TreeMap;
import java.util.stream.Collectors;
import com.knecon.fforesight.service.layoutparser.processor.model.AbstractPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.PageBlockType;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.utils.RectangleTransformations;
import lombok.Getter;
import lombok.Setter;
@@ -21,7 +21,8 @@ import lombok.extern.slf4j.Slf4j;
@Slf4j
public class TablePageBlock extends AbstractPageBlock {
private final TreeMap<CellPosition, Cell> cells = new TreeMap<>();
public static final double CELL_AREA_CONTAINED_THRESHOLD = 0.98;
private final TreeMap<CellPosition, Cell> cellTreeMap = new TreeMap<>();
private final int rotation;
@Getter
@@ -30,10 +31,14 @@ public class TablePageBlock extends AbstractPageBlock {
private int unrotatedRowCount;
private int unrotatedColCount;
private List<List<Cell>> rows;
@Getter
@Setter
private List<Cell> cells;
public TablePageBlock(List<Cell> cells, Rectangle area, int rotation) {
this.cells = cells;
addCells(cells);
minX = area.getLeft();
minY = area.getBottom();
@@ -50,6 +55,7 @@ public class TablePageBlock extends AbstractPageBlock {
return getColCount() == 0 || getRowCount() == 0;
}
public List<List<Cell>> getRows() {
if (rows == null) {
@@ -80,14 +86,17 @@ public class TablePageBlock extends AbstractPageBlock {
public int getColCount() {
return getRows().stream().mapToInt(List::size).max().orElse(0);
return getRows().stream()
.mapToInt(List::size)
.max()
.orElse(0);
}
/**
* Detect header cells (either first row or first column):
* Column is marked as header if cell text is bold and row cell text is not bold.
* Column is marked as header if originalCell text is bold and row originalCell text is not bold.
* Defaults to row.
*/
private void computeHeaders() {
@@ -95,7 +104,7 @@ public class TablePageBlock extends AbstractPageBlock {
if (rows == null) {
rows = computeRows();
}
// A bold cell is a header cell as long as every cell to the left/top is bold, too
// A bold originalCell is a header originalCell as long as every originalCell to the left/top is bold, too
// we move from left to right and top to bottom
for (int rowIndex = 0; rowIndex < rows.size(); rowIndex++) {
List<Cell> rowCells = rows.get(rowIndex);
@@ -120,7 +129,8 @@ public class TablePageBlock extends AbstractPageBlock {
List<Cell> cellsToTheTop = new ArrayList<>();
for (int i = 0; i < rowIndex; i++) {
try {
cellsToTheTop.add(rows.get(i).get(colIndex));
cellsToTheTop.add(rows.get(i)
.get(colIndex));
} catch (IndexOutOfBoundsException e) {
log.debug("No cell {} in row {}, ignoring.", colIndex, rowIndex);
}
@@ -135,7 +145,8 @@ public class TablePageBlock extends AbstractPageBlock {
if (lastHeaderCell != null) {
cell.getHeaderCells().add(lastHeaderCell);
}
if (!cell.getTextBlocks().isEmpty() && cell.getTextBlocks().get(0).getMostPopularWordStyle().equals("bold")) {
if (!cell.getTextBlocks().isEmpty() && cell.getTextBlocks()
.get(0).getMostPopularWordStyle().equals("bold")) {
cell.setHeaderCell(true);
}
}
@@ -151,7 +162,7 @@ public class TablePageBlock extends AbstractPageBlock {
for (int i = 0; i < unrotatedColCount; i++) { // rows
List<Cell> lastRow = new ArrayList<>();
for (int j = unrotatedRowCount - 1; j >= 0; j--) { // cols
Cell cell = cells.get(new CellPosition(j, i));
Cell cell = cellTreeMap.get(new CellPosition(j, i));
if (cell != null) {
lastRow.add(cell);
}
@@ -162,7 +173,7 @@ public class TablePageBlock extends AbstractPageBlock {
for (int i = unrotatedColCount - 1; i >= 0; i--) { // rows
List<Cell> lastRow = new ArrayList<>();
for (int j = 0; j < unrotatedRowCount; j++) { // cols
Cell cell = cells.get(new CellPosition(j, i));
Cell cell = cellTreeMap.get(new CellPosition(j, i));
if (cell != null) {
lastRow.add(cell);
}
@@ -173,7 +184,7 @@ public class TablePageBlock extends AbstractPageBlock {
for (int i = 0; i < unrotatedRowCount; i++) {
List<Cell> lastRow = new ArrayList<>();
for (int j = 0; j < unrotatedColCount; j++) {
Cell cell = cells.get(new CellPosition(i, j)); // JAVA_8 use getOrDefault()
Cell cell = cellTreeMap.get(new CellPosition(i, j)); // JAVA_8 use getOrDefault()
if (cell != null) {
lastRow.add(cell);
}
@@ -187,17 +198,6 @@ public class TablePageBlock extends AbstractPageBlock {
}
private void add(Cell chunk, int row, int col) {
unrotatedRowCount = Math.max(unrotatedRowCount, row + 1);
unrotatedColCount = Math.max(unrotatedColCount, col + 1);
CellPosition cp = new CellPosition(row, col);
cells.put(cp, chunk);
}
private void addCells(List<Cell> cells) {
if (cells.isEmpty()) {
@@ -206,11 +206,12 @@ public class TablePageBlock extends AbstractPageBlock {
cells.removeIf(cell -> cell.getWidth() < 1.1 || cell.getHeight() < 1.1);
List<List<Cell>> rowsOfCells = calculateStructure(cells);
List<List<Cell>> rowsOfCellsMatrix = calculateTableStructure(cells);
for (int i = 0; i < rowsOfCells.size(); i++) {
for (int j = 0; j < rowsOfCells.get(i).size(); j++) {
add(rowsOfCells.get(i).get(j), i, j);
for (int i = 0; i < rowsOfCellsMatrix.size(); i++) {
for (int j = 0; j < rowsOfCellsMatrix.get(i).size(); j++) {
addCellToRowAndCol(rowsOfCellsMatrix.get(i)
.get(j), i, j);
}
}
@@ -221,29 +222,36 @@ public class TablePageBlock extends AbstractPageBlock {
* Calculates the structure of the table. For spanning rows and columns multiple cells with the same values will be inserted.
*
* @param cells The found cells
* @return TablePageBlock Structure
* @return TablePageBlock Structure as a rows of cells matrix
*/
private List<List<Cell>> calculateStructure(List<Cell> cells) {
List<List<Cell>> matrix = new ArrayList<>();
private List<List<Cell>> calculateTableStructure(List<Cell> cells) {
if (cells.isEmpty()) {
return matrix;
return new ArrayList<>();
}
Set<Float> uniqueX = new HashSet<>();
Set<Float> uniqueY = new HashSet<>();
cells.stream().filter(c -> !c.getTextBlocks().isEmpty() || c.getHeight() > 3 && c.getWidth() > 3).forEach(c -> {
uniqueX.add(c.getLeft());
uniqueX.add(c.getRight());
uniqueY.add(c.getBottom());
uniqueY.add(c.getTop());
});
cells.stream()
.filter(c -> !c.getTextBlocks().isEmpty() || c.getHeight() > 3 && c.getWidth() > 3)
.forEach(c -> {
uniqueX.add(c.getLeft());
uniqueX.add(c.getRight());
uniqueY.add(c.getBottom());
uniqueY.add(c.getTop());
});
var sortedUniqueX = uniqueX.stream().sorted().toList();
var sortedUniqueY = uniqueY.stream().sorted().toList();
var sortedUniqueX = uniqueX.stream()
.sorted()
.toList();
var sortedUniqueY = uniqueY.stream()
.sorted()
.toList();
List<List<Cell>> rowsOfCells = new ArrayList<>();
Float prevY = null;
for (Float y : sortedUniqueY) {
List<Cell> row = new ArrayList<>();
@@ -252,44 +260,87 @@ public class TablePageBlock extends AbstractPageBlock {
for (Float x : sortedUniqueX) {
if (prevY != null && prevX != null) {
var cell = new Cell(new Point2D.Float(prevX, prevY), new Point2D.Float(x, y));
var cellFromGridStructure = new Cell(new Point2D.Float(prevX, prevY), new Point2D.Float(x, y));
var intersectionCell = cells.stream().filter(c -> intersects(cell, c)).findFirst();
if (cellFromGridStructure.hasMinimumSize()) {
intersectionCell.ifPresent(value -> cell.getTextBlocks().addAll(value.getTextBlocks()));
if (cell.hasMinimumSize()) {
row.add(cell);
cells.stream()
.map(originalCell -> new CellWithIntersection(originalCell, RectangleTransformations.calculateIntersectedArea(cellFromGridStructure, originalCell)))
.filter(cellWithIntersection -> cellWithIntersection.intersectedArea > 0)
.filter(cellWithIntersection -> cellWithIntersection.originalCell.getArea() > cellWithIntersection.intersectedArea * CELL_AREA_CONTAINED_THRESHOLD)
.max(Comparator.comparing(CellWithIntersection::intersectedArea))
.map(CellWithIntersection::originalCell)
.ifPresent(matchingCell -> cellFromGridStructure.getTextBlocks().addAll(matchingCell.getTextBlocks()));
row.add(cellFromGridStructure);
}
}
prevX = x;
}
if (prevY != null && prevX != null && !row.isEmpty()) {
matrix.add(row);
// exclude empty rows and rows where all text blocks are empty
if (prevY != null && prevX != null && !row.isEmpty() && !row.stream()
.allMatch(cell -> cell.getTextBlocks().isEmpty())) {
rowsOfCells.add(row);
}
prevY = y;
}
Collections.reverse(matrix);
Collections.reverse(rowsOfCells);
return matrix;
}
public boolean intersects(Cell cell1, Cell cell2) {
if (cell1.getHeight() <= 0 || cell2.getHeight() <= 0) {
return false;
// now cells are removed which are part of a column without any text blocks
// this is done by first computing the inverse matrix which contains call columns of cells
// then the column indices that have to be removed are determined
List<List<Cell>> columnsOfCells = new ArrayList<>();
int maxRowLength = rowsOfCells.stream()
.map(List::size)
.max(java.util.Comparator.naturalOrder())
.orElse(0);
for (int i = 0; i < maxRowLength; i++) {
columnsOfCells.add(new ArrayList<>());
}
double x0 = cell1.getX() + 2;
double y0 = cell1.getY() + 2;
return (cell2.x + cell2.width > x0 &&
cell2.y + cell2.height > y0 &&
cell2.x < x0 + cell1.getWidth() -2 &&
cell2.y < y0 + cell1.getHeight() -2);
for (List<Cell> row : rowsOfCells) {
for (int j = 0; j < row.size(); j++) {
columnsOfCells.get(j).add(row.get(j));
}
}
List<Integer> columnIndicesToRemove = new ArrayList<>();
int columnIndex = 0;
for (List<Cell> col : columnsOfCells) {
if (col.stream()
.allMatch(cell -> cell.getTextBlocks().isEmpty())) {
columnIndicesToRemove.add(columnIndex);
}
columnIndex++;
}
columnIndicesToRemove.sort(Collections.reverseOrder());
// update all rows so that the values of the empty columns get removed
var rowsOfCellsBefore = new ArrayList<>(rowsOfCells);
rowsOfCells = new ArrayList<>();
for (List<Cell> row : rowsOfCellsBefore) {
var updatedRow = new ArrayList<>(row);
columnIndicesToRemove.forEach(idxToRemove -> updatedRow.remove(updatedRow.get(idxToRemove)));
rowsOfCells.add(updatedRow);
}
return rowsOfCells;
}
private void addCellToRowAndCol(Cell cell, int row, int col) {
unrotatedRowCount = Math.max(unrotatedRowCount, row + 1);
unrotatedColCount = Math.max(unrotatedColCount, col + 1);
CellPosition cp = new CellPosition(row, col);
cellTreeMap.put(cp, cell);
}
@Override
public String getText() {
@@ -314,7 +365,7 @@ public class TablePageBlock extends AbstractPageBlock {
if (!first) {
sb.append("\n");
}
sb.append('\"').append(textBlock.getText().replaceAll("\"", "\\\"")).append('\"');
sb.append('\"').append(textBlock.getText().replaceAll("\"", "\"")).append('\"');
first = false;
}
}
@@ -328,8 +379,6 @@ public class TablePageBlock extends AbstractPageBlock {
}
public String getTextAsHtml() {
StringBuilder sb = new StringBuilder();
@@ -363,4 +412,9 @@ public class TablePageBlock extends AbstractPageBlock {
return sb.toString();
}
record CellWithIntersection(Cell originalCell, double intersectedArea) {
}
}
@@ -1,8 +0,0 @@
package com.knecon.fforesight.service.layoutparser.processor.model.visualization;
import java.awt.Color;
import java.awt.geom.Line2D;
public record ColoredLine(Line2D line, Color color) {
}
@@ -1,7 +0,0 @@
package com.knecon.fforesight.service.layoutparser.processor.model.visualization;
import java.awt.geom.Point2D;
public record PlacedText(String text, Point2D lineStart) {
}
@@ -0,0 +1,92 @@
package com.knecon.fforesight.service.layoutparser.processor.python_api.adapter;
import java.awt.geom.Rectangle2D;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.ImageType;
import com.knecon.fforesight.service.layoutparser.processor.model.image.ClassifiedImage;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingBox;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResult;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
@Service
@RequiredArgsConstructor
@Slf4j
public class VisualLayoutParsingAdapter {
private static String SIGNATURES = "signature";
public Map<Integer, List<VisualLayoutParsingResult>> buildExtractedTablesPerPage(VisualLayoutParsingResponse visualLayoutParsingResponse) {
Map<Integer, List<VisualLayoutParsingResult>> tableCells = new HashMap<>();
visualLayoutParsingResponse.getData()
.forEach(tableData -> tableCells.computeIfAbsent(tableData.getPage_idx(), tableCell -> new ArrayList<>()).addAll(convertTableCells(tableData.getBoxes())));
return tableCells;
}
public Map<Integer, List<ClassifiedImage>> buildExtractedSignaturesPerPage(VisualLayoutParsingResponse visualLayoutParsingResponse) {
Map<Integer, List<ClassifiedImage>> signatures = new HashMap<>();
if (visualLayoutParsingResponse.getData() != null) {
visualLayoutParsingResponse.getData()
.forEach(tableData -> signatures.computeIfAbsent(tableData.getPage_idx() + 1, tableCell -> new ArrayList<>())
.addAll(convertSignatures(tableData.getPage_idx(), tableData.getBoxes())));
}
return signatures;
}
public List<VisualLayoutParsingResult> convertTableCells(List<VisualLayoutParsingBox> tableObjects) {
List<VisualLayoutParsingResult> parsedTableCells = new ArrayList<>();
tableObjects.stream().forEach(t -> {
VisualLayoutParsingResult result = new VisualLayoutParsingResult();
result.setX0(t.getBox().getX1());
result.setX1(t.getBox().getX2());
result.setY0(t.getBox().getY1());
result.setY1(t.getBox().getY2());
result.setWidth(result.getX1() - result.getX0());
result.setHeight(result.getY1() - result.getY0());
result.setLabel(t.getLabel());
parsedTableCells.add(result);
});
return parsedTableCells;
}
public List<ClassifiedImage> convertSignatures(int pageNumber, List<VisualLayoutParsingBox> tableObjects) {
List<ClassifiedImage> signatures = new ArrayList<>();
tableObjects.stream().forEach(t -> {
if (t.getLabel().equals(SIGNATURES)) {
ClassifiedImage signature = new ClassifiedImage(new Rectangle2D.Float(t.getBox().getX1(),
t.getBox().getY1(),
t.getBox().getX2() - t.getBox().getX1(),
t.getBox().getY2() - t.getBox().getY1()), ImageType.SIGNATURE, true, false, false, pageNumber);
signatures.add(signature);
}
});
return signatures;
}
}
@@ -0,0 +1,18 @@
package com.knecon.fforesight.service.layoutparser.processor.python_api.model.table;
import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Data;
import lombok.NoArgsConstructor;
@Data
@Builder
@NoArgsConstructor
@AllArgsConstructor
public class VisualLayoutParsingBox {
private VisualLayoutParsingBoxValue box;
private String label;
private float probability;
}
@@ -0,0 +1,19 @@
package com.knecon.fforesight.service.layoutparser.processor.python_api.model.table;
import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Data;
import lombok.NoArgsConstructor;
@Data
@Builder
@NoArgsConstructor
@AllArgsConstructor
public class VisualLayoutParsingBoxValue {
private float x1;
private float y1;
private float x2;
private float y2;
}
@@ -0,0 +1,21 @@
package com.knecon.fforesight.service.layoutparser.processor.python_api.model.table;
import java.util.List;
import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Data;
import lombok.NoArgsConstructor;
@Data
@Builder
@NoArgsConstructor
@AllArgsConstructor
public class VisualLayoutParsingData {
private int page_idx;
private List<VisualLayoutParsingBox> boxes;
}
@@ -0,0 +1,23 @@
package com.knecon.fforesight.service.layoutparser.processor.python_api.model.table;
import java.util.List;
import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Data;
import lombok.NoArgsConstructor;
@Data
@Builder
@NoArgsConstructor
@AllArgsConstructor
public class VisualLayoutParsingResponse {
private String dossierId;
private String fileId;
private String targetFileExtension;
private String responseFileExtension;
private String X_TENANT_ID;
private List<VisualLayoutParsingData> data;
}
@@ -0,0 +1,22 @@
package com.knecon.fforesight.service.layoutparser.processor.python_api.model.table;
import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Data;
import lombok.NoArgsConstructor;
@Data
@Builder
@NoArgsConstructor
@AllArgsConstructor
public class VisualLayoutParsingResult {
private float x0;
private float y0;
private float x1;
private float y1;
private float width;
private float height;
private String label;
}
@@ -1,21 +1,21 @@
package com.knecon.fforesight.service.layoutparser.processor.services;
import java.awt.geom.Line2D;
import static com.knecon.fforesight.service.layoutparser.processor.utils.GeometricComparators.X_FIRST_RULING_COMPARATOR;
import java.awt.geom.Point2D;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.stream.Collectors;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.model.table.CleanRulings;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Rectangle;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Ruling;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableCells;
import com.knecon.fforesight.service.layoutparser.processor.utils.DoubleComparisons;
import com.knecon.fforesight.service.layoutparser.processor.utils.UnionFind;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
@@ -25,25 +25,145 @@ import lombok.extern.slf4j.Slf4j;
@RequiredArgsConstructor
public class RulingCleaningService {
private static final float THRESHOLD = 6;
private static final float THRESHOLD_X_VERTICAL = 1;
private static final float THRESHOLD_Y_VERTICAL = 2;
private static final float THRESHOLD_X_HORIZONTAL = 2;
private static final float THRESHOLD_Y_HORIZONTAL = 3;
public CleanRulings getCleanRulings(List<TableCells> tableCells, List<Ruling> rulings) {
Rulings verticalAndHorizontalRulingLines;
if (!rulings.isEmpty()) {
snapPoints(rulings);
verticalAndHorizontalRulingLines = extractVerticalAndHorizontalRulingLines(rulings);
} else {
verticalAndHorizontalRulingLines = getRulingsFromParsedCells(tableCells);
}
verticalAndHorizontalRulingLines.verticalLines.sort(X_FIRST_RULING_COMPARATOR);
verticalAndHorizontalRulingLines.horizontalLines.sort(X_FIRST_RULING_COMPARATOR);
verticalAndHorizontalRulingLines = cleanRulings(verticalAndHorizontalRulingLines);
return CleanRulings.builder().vertical(verticalAndHorizontalRulingLines.verticalLines()).horizontal(verticalAndHorizontalRulingLines.horizontalLines()).build();
}
private Rulings cleanRulings(Rulings rulings) {
List<List<Rectangle>> groupedOverlappingVerticalRectangles = groupOverlappingRectangles(rulings.verticalLines.stream()
.map(RulingCleaningService::getOverlapRectangle)
.distinct()
.toList());
List<Ruling> cleanedVerticalRulings = groupedOverlappingVerticalRectangles.stream()
.map(rectList -> getXCenteredRuling(Rectangle.boundingBoxOf(rectList)))
.toList();
List<List<Rectangle>> groupedOverlappingHorizontalRectangles = groupOverlappingRectangles(rulings.horizontalLines.stream()
.map(RulingCleaningService::getOverlapRectangle)
.distinct()
.toList());
List<Ruling> cleanedHorizontalRulings = groupedOverlappingHorizontalRectangles.stream()
.map(rectList -> getYCenteredRuling(Rectangle.boundingBoxOf(rectList)))
.collect(Collectors.toList());
return new Rulings(cleanedVerticalRulings, cleanedHorizontalRulings);
}
private List<List<Rectangle>> groupOverlappingRectangles(List<Rectangle> rectangles) {
UnionFind<Rectangle> unionFind = new UnionFind<>();
for (int i = 0; i < rectangles.size(); i++) {
for (int j = i + 1; j < rectangles.size(); j++) {
Rectangle rectangle1 = rectangles.get(i);
Rectangle rectangle2 = rectangles.get(j);
// we can stop early when we are too far off because of x-y-sorting
if(rectangle1.getRight() < rectangle2.getLeft() && rectangle1.getBottom() < rectangle2.getTop()) {
break;
}
if (rectangle1.intersects(rectangle2)) {
unionFind.union(rectangle1, rectangle2);
}
}
}
Map<Rectangle, List<Rectangle>> groups = new HashMap<>();
for (Rectangle rectangle : rectangles) {
Rectangle root = unionFind.find(rectangle);
groups.computeIfAbsent(root, k -> new ArrayList<>()).add(rectangle);
}
return new ArrayList<>(groups.values());
}
private static Rectangle getOverlapRectangle(Ruling ruling) {
float top;
float left;
float w;
float h;
if (ruling.x1 < ruling.x2) {
left = ruling.x1;
w = ruling.x2 - ruling.x1;
} else {
left = ruling.x2;
w = ruling.x1 - ruling.x2;
}
if (ruling.y1 < ruling.y2) {
top = ruling.y1;
h = ruling.y2 - ruling.y1;
} else {
top = ruling.y2;
h = ruling.y1 - ruling.y2;
}
if (ruling.horizontal()) {
return new Rectangle(top - THRESHOLD_Y_HORIZONTAL, left - THRESHOLD_X_HORIZONTAL, w + 2 * THRESHOLD_X_HORIZONTAL, h + 2 * THRESHOLD_Y_HORIZONTAL);
} else {
return new Rectangle(top - THRESHOLD_Y_VERTICAL, left - THRESHOLD_X_VERTICAL, w + 2 * THRESHOLD_X_VERTICAL, h + 2 * THRESHOLD_Y_VERTICAL);
}
}
public static Ruling getXCenteredRuling(Rectangle rectangle) {
float x = (float) rectangle.getCenterX();
float y1 = rectangle.getTop();
float y2 = rectangle.getBottom();
Point2D point1 = new Point2D.Float(x, y1 + THRESHOLD_Y_VERTICAL);
Point2D point2 = new Point2D.Float(x, y2 - THRESHOLD_Y_VERTICAL);
return new Ruling(point1, point2);
}
public static Ruling getYCenteredRuling(Rectangle rectangle) {
float x1 = rectangle.getLeft();
float x2 = rectangle.getRight();
float y = (float) rectangle.getCenterY();
Point2D point1 = new Point2D.Float(x1 + THRESHOLD_X_HORIZONTAL, y);
Point2D point2 = new Point2D.Float(x2 - THRESHOLD_X_HORIZONTAL, y);
return new Ruling(point1, point2);
}
private Rulings extractVerticalAndHorizontalRulingLines(List<Ruling> rulings) {
List<Ruling> vrs = new ArrayList<>();
for (Ruling vr : rulings) {
if (vr.vertical()) {
vrs.add(vr);
}
}
if (vrs.isEmpty()) {
vrs.addAll(extractVerticalRulings(tableCells));
}
List<Ruling> verticalRulingLines = collapseOrientedRulings(vrs);
List<Ruling> hrs = new ArrayList<>();
for (Ruling hr : rulings) {
@@ -51,98 +171,26 @@ public class RulingCleaningService {
hrs.add(hr);
}
}
if (hrs.isEmpty()) {
hrs.addAll(extractHorizontalRulings(tableCells));
}
List<Ruling> horizontalRulingLines = collapseOrientedRulings(hrs);
return CleanRulings.builder().vertical(verticalRulingLines).horizontal(horizontalRulingLines).build();
return new Rulings(vrs, hrs);
}
public void snapPoints(List<? extends Line2D.Float> rulings) {
private Rulings getRulingsFromParsedCells(List<TableCells> tableCells) {
// collect points and keep a Line -> p1,p2 map
Map<Line2D.Float, Point2D[]> linesToPoints = new HashMap<>();
List<Point2D> points = new ArrayList<>();
for (Line2D.Float r : rulings) {
Point2D p1 = r.getP1();
Point2D p2 = r.getP2();
linesToPoints.put(r, new Point2D[]{p1, p2});
points.add(p1);
points.add(p2);
}
// snap by X
points.sort(Comparator.comparingDouble(Point2D::getX));
List<List<Point2D>> groupedPoints = new ArrayList<>();
groupedPoints.add(new ArrayList<>(Collections.singletonList(points.get(0))));
for (Point2D p : points.subList(1, points.size() - 1)) {
List<Point2D> last = groupedPoints.get(groupedPoints.size() - 1);
if (Math.abs(p.getX() - last.get(0).getX()) < THRESHOLD) {
groupedPoints.get(groupedPoints.size() - 1).add(p);
} else {
groupedPoints.add(new ArrayList<>(Collections.singletonList(p)));
}
}
for (List<Point2D> group : groupedPoints) {
float avgLoc = 0;
for (Point2D p : group) {
avgLoc += p.getX();
}
avgLoc /= group.size();
for (Point2D p : group) {
p.setLocation(avgLoc, p.getY());
}
}
// ---
// snap by Y
points.sort(Comparator.comparingDouble(Point2D::getY));
groupedPoints = new ArrayList<>();
groupedPoints.add(new ArrayList<>(Collections.singletonList(points.get(0))));
for (Point2D p : points.subList(1, points.size() - 1)) {
List<Point2D> last = groupedPoints.get(groupedPoints.size() - 1);
if (Math.abs(p.getY() - last.get(0).getY()) < THRESHOLD) {
groupedPoints.get(groupedPoints.size() - 1).add(p);
} else {
groupedPoints.add(new ArrayList<>(Collections.singletonList(p)));
}
}
for (List<Point2D> group : groupedPoints) {
float avgLoc = 0;
for (Point2D p : group) {
avgLoc += p.getY();
}
avgLoc /= group.size();
for (Point2D p : group) {
p.setLocation(p.getX(), avgLoc);
}
}
// ---
// finally, modify lines
for (Map.Entry<Line2D.Float, Point2D[]> ltp : linesToPoints.entrySet()) {
Point2D[] p = ltp.getValue();
ltp.getKey().setLine(p[0], p[1]);
}
List<Ruling> vrs = extractVerticalRulingsFromParsedCells(tableCells);
List<Ruling> hrs = extractHorizontalRulingsFromParsedCells(tableCells);
return new Rulings(vrs, hrs);
}
private Collection<? extends Ruling> extractVerticalRulings(List<TableCells> cvParsedTableCells) {
private List<Ruling> extractVerticalRulingsFromParsedCells(List<TableCells> tableCells) {
List<Ruling> vrs = new ArrayList<>();
if (cvParsedTableCells != null) {
for (TableCells cvParsedTableCell : cvParsedTableCells) {
Ruling leftLine = createRuling(cvParsedTableCell.getX0(), cvParsedTableCell.getX0(), cvParsedTableCell.getY0(), cvParsedTableCell.getY1());
Ruling rightLine = createRuling(cvParsedTableCell.getX1(), cvParsedTableCell.getX1(), cvParsedTableCell.getY0(), cvParsedTableCell.getY1());
if (tableCells != null) {
for (TableCells tableCell : tableCells) {
Ruling leftLine = createRuling(tableCell.getX0(), tableCell.getX0(), tableCell.getY0(), tableCell.getY1());
Ruling rightLine = createRuling(tableCell.getX1(), tableCell.getX1(), tableCell.getY0(), tableCell.getY1());
vrs.add(leftLine);
vrs.add(rightLine);
}
@@ -151,19 +199,18 @@ public class RulingCleaningService {
}
private Collection<? extends Ruling> extractHorizontalRulings(List<TableCells> cvParsedTableCells) {
private List<Ruling> extractHorizontalRulingsFromParsedCells(List<TableCells> tableCells) {
List<Ruling> hrs = new ArrayList<>();
if (cvParsedTableCells != null) {
for (TableCells cvParsedTableCell : cvParsedTableCells) {
Ruling topLine = createRuling(cvParsedTableCell.getX0(), cvParsedTableCell.getX1(), cvParsedTableCell.getY1(), cvParsedTableCell.getY1());
Ruling baseLine = createRuling(cvParsedTableCell.getX0(), cvParsedTableCell.getX1(), cvParsedTableCell.getY0(), cvParsedTableCell.getY0());
if (tableCells != null) {
for (TableCells tableCell : tableCells) {
Ruling topLine = createRuling(tableCell.getX0(), tableCell.getX1(), tableCell.getY1(), tableCell.getY1());
Ruling baseLine = createRuling(tableCell.getX0(), tableCell.getX1(), tableCell.getY0(), tableCell.getY0());
hrs.add(topLine);
hrs.add(baseLine);
}
}
return hrs;
}
@@ -189,46 +236,8 @@ public class RulingCleaningService {
}
private List<Ruling> collapseOrientedRulings(List<Ruling> lines) {
private record Rulings(List<Ruling> verticalLines, List<Ruling> horizontalLines) {
int COLINEAR_OR_PARALLEL_PIXEL_EXPAND_AMOUNT = 1;
return collapseOrientedRulings(lines, COLINEAR_OR_PARALLEL_PIXEL_EXPAND_AMOUNT);
}
private List<Ruling> collapseOrientedRulings(List<Ruling> lines, int expandAmount) {
ArrayList<Ruling> rv = new ArrayList<>();
lines.sort((a, b) -> {
final float diff = a.getPosition() - b.getPosition();
return Float.compare(diff == 0 ? a.getStart() - b.getStart() : diff, 0f);
});
for (Ruling next_line : lines) {
Ruling last = rv.isEmpty() ? null : rv.get(rv.size() - 1);
// if current line colinear with next, and are "close enough": expand current line
if (last != null && DoubleComparisons.feq(next_line.getPosition(), last.getPosition()) && last.nearlyIntersects(next_line, expandAmount)) {
final float lastStart = last.getStart();
final float lastEnd = last.getEnd();
final boolean lastFlipped = lastStart > lastEnd;
final boolean nextFlipped = next_line.getStart() > next_line.getEnd();
boolean differentDirections = nextFlipped != lastFlipped;
float nextS = differentDirections ? next_line.getEnd() : next_line.getStart();
float nextE = differentDirections ? next_line.getStart() : next_line.getEnd();
final float newStart = lastFlipped ? Math.max(nextS, lastStart) : Math.min(nextS, lastStart);
final float newEnd = lastFlipped ? Math.min(nextE, lastEnd) : Math.max(nextE, lastEnd);
last.setStartEnd(newStart, newEnd);
assert !last.oblique();
} else if (next_line.length() == 0) {
continue;
} else {
rv.add(next_line);
}
}
return rv;
}
}
@@ -1,14 +1,13 @@
package com.knecon.fforesight.service.layoutparser.processor.services;
import java.awt.geom.Point2D;
import static com.knecon.fforesight.service.layoutparser.processor.utils.GeometricComparators.CELL_SIZE_COMPARATOR;
import static com.knecon.fforesight.service.layoutparser.processor.utils.GeometricComparators.RECTANGLE_SIZE_COMPARATOR;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.HashMap;
import java.util.HashSet;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.stream.Collectors;
import org.springframework.stereotype.Service;
@@ -21,59 +20,15 @@ import com.knecon.fforesight.service.layoutparser.processor.model.table.Ruling;
import com.knecon.fforesight.service.layoutparser.processor.model.table.TablePageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.utils.DoubleComparisons;
import com.knecon.fforesight.service.layoutparser.processor.utils.RectangularIntersectionFinder;
import com.knecon.fforesight.service.layoutparser.processor.utils.SpreadsheetFinder;
@Service
public class TableExtractionService {
private static final Comparator<Point2D> X_FIRST_POINT_COMPARATOR = (arg0, arg1) -> {
int rv = 0;
float arg0X = DoubleComparisons.round(arg0.getX(), 2);
float arg0Y = DoubleComparisons.round(arg0.getY(), 2);
float arg1X = DoubleComparisons.round(arg1.getX(), 2);
float arg1Y = DoubleComparisons.round(arg1.getY(), 2);
if (arg0X > arg1X) {
rv = 1;
} else if (arg0X < arg1X) {
rv = -1;
} else if (arg0Y > arg1Y) {
rv = 1;
} else if (arg0Y < arg1Y) {
rv = -1;
}
return rv;
};
private static final Comparator<Point2D> POINT_COMPARATOR = (arg0, arg1) -> {
int rv = 0;
float arg0X = DoubleComparisons.round(arg0.getX(), 2);
float arg0Y = DoubleComparisons.round(arg0.getY(), 2);
float arg1X = DoubleComparisons.round(arg1.getX(), 2);
float arg1Y = DoubleComparisons.round(arg1.getY(), 2);
if (arg0Y > arg1Y) {
rv = 1;
} else if (arg0Y < arg1Y) {
rv = -1;
} else if (arg0X > arg1X) {
rv = 1;
} else if (arg0X < arg1X) {
rv = -1;
}
return rv;
};
public boolean contains(Cell cell, double x, double y, double w, double h) {
if (cell.isEmpty() || w <= 0 || h <= 0) {
return false;
}
double x0 = cell.getX();
double y0 = cell.getY();
return (x >= x0 - 2 && y >= y0 - 2 && (x + w) <= x0 + cell.getWidth() + 2 && (y + h) <= y0 + cell.getHeight() + 2);
}
private static final int MAX_TABLE_CONTAINED_CELLS_WITH_TEXT = 1;
private static final int TEXT_BLOCK_CONTAINMENT_TOLERANCE = 2;
private static final double TABLE_UNIFORMITY_THRESHOLD = 0.7;
/**
@@ -89,22 +44,18 @@ public class TableExtractionService {
* @param cleanRulings The lines used to build the table.
* @param page Page object that contains textblocks and statistics.
*/
public void extractTables(CleanRulings cleanRulings, ClassificationPage page) {
List<Cell> cells = findCells(cleanRulings.getHorizontal(), cleanRulings.getVertical());
List<TextPageBlock> toBeRemoved = new ArrayList<>();
// sort cells by size (height * width) ascending so that textBlocks are always assigned to the smallest cells that contain them
cells.sort(CELL_SIZE_COMPARATOR);
for (AbstractPageBlock abstractPageBlock : page.getTextBlocks()) {
TextPageBlock textBlock = (TextPageBlock) abstractPageBlock;
for (Cell cell : cells) {
if (cell.hasMinimumSize() && contains(cell,
textBlock.getPdfMinX(),
textBlock.getPdfMinY(),
textBlock.getPdfMaxX() - textBlock.getPdfMinX(),
textBlock.getPdfMaxY() - textBlock.getPdfMinY())) {
if (cell.hasMinimumSize() && doesCellContainTextBlock(cell, textBlock)) {
cell.addTextBlock(textBlock);
toBeRemoved.add(textBlock);
break;
}
}
@@ -113,245 +64,94 @@ public class TableExtractionService {
cells = new ArrayList<>(new HashSet<>(cells));
DoubleComparisons.sort(cells, Rectangle.ILL_DEFINED_ORDER);
List<Rectangle> spreadsheetAreas = findSpreadsheetsFromCells(cells);
List<Rectangle> spreadsheetAreas = SpreadsheetFinder.findSpreadsheetsFromCells(cells);
// sort spreadsheetAreas by size (height * width) ascending so that cells are placed in the smallest tables first
// this way no cell duplication occurs when tables are contained in other tables and only the most inner table contains the cells
spreadsheetAreas.sort(RECTANGLE_SIZE_COMPARATOR);
List<TablePageBlock> tables = new ArrayList<>();
for (Rectangle area : spreadsheetAreas) {
List<Cell> overlappingCells = new ArrayList<>();
List<Cell> containedCells = new ArrayList<>();
for (Cell c : cells) {
if (c.hasMinimumSize() && c.intersects(area)) {
overlappingCells.add(c);
if (c.hasMinimumSize() && area.contains(c)) {
containedCells.add(c);
}
}
tables.add(new TablePageBlock(overlappingCells, area, page.getRotation()));
var containedCellsWithText = containedCells.stream()
.filter(cell -> !cell.getTextBlocks().isEmpty())
.toList();
// verify if table would contain fewer cells with text than the threshold allows
if (containedCellsWithText.size() >= MAX_TABLE_CONTAINED_CELLS_WITH_TEXT && checkIfTableCellsAreUniform(containedCells)) {
tables.add(new TablePageBlock(containedCells, area, page.getRotation()));
cells.removeAll(containedCells);
}
}
for (TablePageBlock table : tables) {
int position = -1;
Iterator<AbstractPageBlock> itty = page.getTextBlocks().iterator();
while (itty.hasNext()) {
AbstractPageBlock textBlock = itty.next();
if (textBlock instanceof TextPageBlock ? table.containsBlock((TextPageBlock) textBlock) : table.contains(textBlock) && position == -1) {
position = page.getTextBlocks().indexOf(textBlock);
for (AbstractPageBlock pageBlock : page.getTextBlocks()) {
if (pageBlock instanceof TextPageBlock ? table.containsBlock((TextPageBlock) pageBlock) : table.contains(pageBlock) && position == -1) {
position = page.getTextBlocks().indexOf(pageBlock);
}
}
if (position != -1) {
page.getTextBlocks().add(position, table);
var toBeRemoved = table.getCells()
.stream()
.map(Cell::getTextBlocks)
.flatMap(List::stream)
.toList();
// remove text blocks from the page that were also added with the table (from its contained cells)
page.getTextBlocks().removeAll(toBeRemoved);
}
}
page.getTextBlocks().removeAll(toBeRemoved);
}
public List<Cell> findCells(List<Ruling> horizontalRulingLines, List<Ruling> verticalRulingLines) {
private boolean checkIfTableCellsAreUniform(List<Cell> containedCells) {
// Fix for 211.pdf
for (Ruling r : horizontalRulingLines) {
if (r.getX2() < r.getX1()) {
double a = r.getX2();
r.x2 = (float) r.getX1();
r.x1 = (float) a;
}
if(containedCells.size() <= 2) {
return true;
}
List<Cell> cellsFound = new ArrayList<>();
Map<Point2D, Ruling[]> intersectionPoints = Ruling.findIntersections(horizontalRulingLines, verticalRulingLines);
List<Point2D> intersectionPointsList = new ArrayList<>(intersectionPoints.keySet());
intersectionPointsList.sort(POINT_COMPARATOR);
Map<Long, List<Long>> cellsGroupedByRoundedWidth = containedCells.stream()
.map(Rectangle::getWidth)
.map(size -> Math.round(size / 10.0) * 10)
.collect(Collectors.groupingBy(Long::longValue));
for (int i = 0; i < intersectionPointsList.size(); i++) {
Point2D topLeft = intersectionPointsList.get(i);
Ruling[] hv = intersectionPoints.get(topLeft);
// CrossingPointsDirectlyBelow( topLeft );
List<Point2D> xPoints = new ArrayList<>();
// CrossingPointsDirectlyToTheRight( topLeft );
List<Point2D> yPoints = new ArrayList<>();
for (Point2D p : intersectionPointsList.subList(i, intersectionPointsList.size())) {
if (p.getX() == topLeft.getX() && p.getY() > topLeft.getY()) {
xPoints.add(p);
}
if (p.getY() == topLeft.getY() && p.getX() > topLeft.getX()) {
yPoints.add(p);
}
}
outer:
for (Point2D xPoint : xPoints) {
// is there a vertical edge b/w topLeft and xPoint?
if (!hv[1].equals(intersectionPoints.get(xPoint)[1])) {
continue;
}
for (Point2D yPoint : yPoints) {
// is there an horizontal edge b/w topLeft and yPoint ?
if (!hv[0].equals(intersectionPoints.get(yPoint)[0])) {
continue;
}
Point2D btmRight = new Point2D.Float((float) yPoint.getX(), (float) xPoint.getY());
if (intersectionPoints.containsKey(btmRight) && intersectionPoints.get(btmRight)[0].equals(intersectionPoints.get(xPoint)[0]) && intersectionPoints.get(btmRight)[1].equals(
intersectionPoints.get(yPoint)[1])) {
cellsFound.add(new Cell(topLeft, btmRight));
break outer;
}
}
}
}
// TODO create cells for vertical ruling lines with aligned endpoints at the top/bottom of a grid
// that aren't connected with an horizontal ruler?
// see: https://github.com/jazzido/tabula-extractor/issues/78#issuecomment-41481207
return cellsFound;
return (double) cellsGroupedByRoundedWidth.size() / containedCells.size() <= TABLE_UNIFORMITY_THRESHOLD;
}
private List<Rectangle> findSpreadsheetsFromCells(List<? extends Rectangle> cells) {
// via: http://stackoverflow.com/questions/13746284/merging-multiple-adjacent-rectangles-into-one-polygon
List<Rectangle> rectangles = new ArrayList<>();
Set<Point2D> pointSet = new HashSet<>();
Map<Point2D, Point2D> edgesH = new HashMap<>();
Map<Point2D, Point2D> edgesV = new HashMap<>();
int i = 0;
private boolean doesCellContainTextBlock(Cell cell, TextPageBlock textBlock) {
for (Rectangle cell : cells) {
for (Point2D pt : cell.getPoints()) {
if (pointSet.contains(pt)) { // shared vertex, remove it
pointSet.remove(pt);
} else {
pointSet.add(pt);
}
}
double x = textBlock.getPdfMinX();
double y = textBlock.getPdfMinY();
double w = textBlock.getPdfMaxX() - textBlock.getPdfMinX();
double h = textBlock.getPdfMaxY() - textBlock.getPdfMinY();
if (cell.isEmpty() || w <= 0 || h <= 0) {
return false;
}
// X first sort
List<Point2D> pointsSortX = new ArrayList<>(pointSet);
pointsSortX.sort(X_FIRST_POINT_COMPARATOR);
// Y first sort
List<Point2D> pointsSortY = new ArrayList<>(pointSet);
pointsSortY.sort(POINT_COMPARATOR);
while (i < pointSet.size()) {
float currY = (float) pointsSortY.get(i).getY();
while (i < pointSet.size() && DoubleComparisons.feq(pointsSortY.get(i).getY(), currY)) {
edgesH.put(pointsSortY.get(i), pointsSortY.get(i + 1));
edgesH.put(pointsSortY.get(i + 1), pointsSortY.get(i));
i += 2;
}
}
i = 0;
while (i < pointSet.size()) {
float currX = (float) pointsSortX.get(i).getX();
while (i < pointSet.size() && DoubleComparisons.feq(pointsSortX.get(i).getX(), currX)) {
edgesV.put(pointsSortX.get(i), pointsSortX.get(i + 1));
edgesV.put(pointsSortX.get(i + 1), pointsSortX.get(i));
i += 2;
}
}
// Get all the polygons
List<List<PolygonVertex>> polygons = new ArrayList<>();
Point2D nextVertex;
while (!edgesH.isEmpty()) {
ArrayList<PolygonVertex> polygon = new ArrayList<>();
Point2D first = edgesH.keySet().iterator().next();
polygon.add(new PolygonVertex(first, Direction.HORIZONTAL));
edgesH.remove(first);
while (true) {
PolygonVertex curr = polygon.get(polygon.size() - 1);
PolygonVertex lastAddedVertex;
if (curr.direction == Direction.HORIZONTAL) {
nextVertex = edgesV.get(curr.point);
edgesV.remove(curr.point);
lastAddedVertex = new PolygonVertex(nextVertex, Direction.VERTICAL);
} else {
nextVertex = edgesH.get(curr.point);
edgesH.remove(curr.point);
lastAddedVertex = new PolygonVertex(nextVertex, Direction.HORIZONTAL);
}
polygon.add(lastAddedVertex);
if (lastAddedVertex.equals(polygon.get(0))) {
// closed polygon
polygon.remove(polygon.size() - 1);
break;
}
}
for (PolygonVertex vertex : polygon) {
edgesH.remove(vertex.point);
edgesV.remove(vertex.point);
}
polygons.add(polygon);
}
// calculate grid-aligned minimum area rectangles for each found polygon
for (List<PolygonVertex> poly : polygons) {
float top = Float.MAX_VALUE;
float left = Float.MAX_VALUE;
float bottom = Float.MIN_VALUE;
float right = Float.MIN_VALUE;
for (PolygonVertex pt : poly) {
top = (float) Math.min(top, pt.point.getY());
left = (float) Math.min(left, pt.point.getX());
bottom = (float) Math.max(bottom, pt.point.getY());
right = (float) Math.max(right, pt.point.getX());
}
rectangles.add(new Rectangle(top, left, right - left, bottom - top));
}
return rectangles;
double x0 = cell.getX();
double y0 = cell.getY();
return (x >= x0 - TEXT_BLOCK_CONTAINMENT_TOLERANCE
&& y >= y0 - TEXT_BLOCK_CONTAINMENT_TOLERANCE
&& (x + w) <= x0 + cell.getWidth() + 2 * TEXT_BLOCK_CONTAINMENT_TOLERANCE
&& (y + h) <= y0 + cell.getHeight() + 2 * TEXT_BLOCK_CONTAINMENT_TOLERANCE);
}
private enum Direction {
HORIZONTAL,
VERTICAL
}
static class PolygonVertex {
Point2D point;
Direction direction;
PolygonVertex(Point2D point, Direction direction) {
this.direction = direction;
this.point = point;
}
@Override
public boolean equals(Object other) {
if (this == other) {
return true;
}
if (!(other instanceof PolygonVertex)) {
return false;
}
return this.point.equals(((PolygonVertex) other).point);
}
@Override
public int hashCode() {
return this.point.hashCode();
}
@Override
public String toString() {
return String.format("%s[point=%s,direction=%s]", this.getClass().getName(), this.point.toString(), this.direction.toString());
}
public static List<Cell> findCells(List<Ruling> horizontalRulingLines, List<Ruling> verticalRulingLines) {
return RectangularIntersectionFinder.find(horizontalRulingLines, verticalRulingLines)
.stream()
.map(Cell::new)
.collect(Collectors.toList());
}
}
@@ -7,19 +7,19 @@ import static java.util.stream.Collectors.toList;
import java.awt.geom.Rectangle2D;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.HashSet;
import java.util.LinkedList;
import java.util.List;
import java.util.Map;
import java.util.NoSuchElementException;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.processor.model.AbstractPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationDocument;
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationFooter;
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationHeader;
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationPage;
import com.knecon.fforesight.service.layoutparser.processor.model.image.ClassifiedImage;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.DocumentTree;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Footer;
@@ -31,6 +31,8 @@ import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Pa
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Paragraph;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Section;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.AtomicTextBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.image.ClassifiedImage;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.utils.IdBuilder;
import com.knecon.fforesight.service.layoutparser.processor.utils.TextPositionOperations;
@@ -95,18 +97,21 @@ public class DocumentGraphFactory {
Rectangle2D position = image.getPosition();
Page page = context.getPage(image.getPage());
Image imageNode = Image.builder()
var imageBuilder = Image.builder()
.id(IdBuilder.buildId(Set.of(page), List.of(position)))
.imageType(image.getImageType())
.position(position)
.transparent(image.isHasTransparency())
.page(page)
.documentTree(context.getDocumentTree())
.build();
.documentTree(context.getDocumentTree());
if (image.isSourceByAi()) {
imageBuilder.engines(new HashSet<>(Set.of(LayoutEngine.AI)));
}
Image imageNode = imageBuilder.build();
page.getMainBody().add(imageNode);
List<Integer> tocId = context.getDocumentTree().createNewChildEntryAndReturnId(section, imageNode);
imageNode.setTreeId(tocId);
List<Integer> treeId = context.getDocumentTree().createNewChildEntryAndReturnId(section, imageNode);
imageNode.setTreeId(treeId);
}
@@ -80,6 +80,7 @@ public class DocumentDataMapper {
.treeId(toPrimitiveIntArray(entry.getTreeId()))
.children(entry.getChildren().stream().map(DocumentDataMapper::toEntryData).toList())
.type(entry.getType())
.engines(entry.getNode().getEngines())
.atomicBlockIds(atomicTextBlocks)
.pageNumbers(entry.getNode().getPages().stream().map(Page::getNumber).map(Integer::longValue).toArray(Long[]::new))
.properties(properties)
@@ -1,12 +1,11 @@
package com.knecon.fforesight.service.layoutparser.processor.services.visualization;
import static com.knecon.fforesight.service.layoutparser.processor.services.visualization.ViewerDocumentService.LINE_WIDTH;
import java.awt.Color;
import java.awt.geom.Line2D;
import java.awt.geom.Point2D;
import java.awt.geom.Rectangle2D;
import java.awt.geom.RectangularShape;
import java.io.File;
import java.util.ArrayList;
import java.util.Collections;
import java.util.Comparator;
@@ -18,55 +17,100 @@ import java.util.stream.Stream;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Page;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.SemanticNode;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Table;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.TableCell;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.ColoredLine;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.ColoredRectangle;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.LayoutGrid;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.PlacedText;
import com.knecon.fforesight.service.layoutparser.processor.utils.RectangleTransformations;
import com.knecon.fforesight.service.viewerdoc.ContentStreams;
import com.knecon.fforesight.service.viewerdoc.model.ColoredLine;
import com.knecon.fforesight.service.viewerdoc.model.ColoredRectangle;
import com.knecon.fforesight.service.viewerdoc.model.LayoutGrid;
import com.knecon.fforesight.service.viewerdoc.model.PlacedText;
import com.knecon.fforesight.service.viewerdoc.model.Standard14EmbeddableFont;
import com.knecon.fforesight.service.viewerdoc.model.Visualizations;
import com.knecon.fforesight.service.viewerdoc.service.ViewerDocumentService;
import io.micrometer.observation.annotation.Observed;
import lombok.AccessLevel;
import lombok.RequiredArgsConstructor;
import lombok.SneakyThrows;
import lombok.experimental.FieldDefaults;
@Service
@RequiredArgsConstructor
@FieldDefaults(level = AccessLevel.PRIVATE, makeFinal = true)
public class LayoutGridService {
private static final Color INNER_LINES_COLOR = new Color(255, 175, 175);
private static final Color PARAGRAPH_COLOR = new Color(70, 130, 180);
public static final Color TABLE_COLOR = new Color(102, 205, 170);
public static final Color SECTION_COLOR = new Color(50, 50, 50);
public static final Color HEADLINE_COLOR = new Color(162, 56, 56);
public static final Color HEADER_COLOR = new Color(171, 131, 6);
public static final Color IMAGE_COLOR = new Color(253, 63, 146);
ViewerDocumentService viewerDocumentService;
static float FONT_SIZE = 10f;
static float LINE_WIDTH = 1f;
static Standard14EmbeddableFont FONT = Standard14EmbeddableFont.helvetica();
static Color INNER_LINES_COLOR = new Color(255, 175, 175);
static Color PARAGRAPH_COLOR = new Color(70, 130, 180);
static Color TABLE_COLOR = new Color(102, 205, 170);
static Color SECTION_COLOR = new Color(50, 50, 50);
static Color HEADLINE_COLOR = new Color(162, 56, 56);
static Color HEADER_COLOR = new Color(171, 131, 6);
static Color IMAGE_COLOR = new Color(253, 63, 146);
public LayoutGrid createLayoutGrid(Document document) {
@SneakyThrows
@Observed(name = "ViewerDocumentService", contextualName = "create-viewer-document")
public void addLayoutGrid(File originFile, Document document, File destinationFile, boolean layerVisibilityDefaultValue) {
Visualizations layoutGrid = this.addLayoutGrid(document, layerVisibilityDefaultValue, false);
Visualizations visualLayoutGrid = this.addLayoutGrid(document, layerVisibilityDefaultValue, true);
viewerDocumentService.addVisualizationsOnPage(originFile, destinationFile, List.of(layoutGrid, visualLayoutGrid));
}
@SneakyThrows
@Observed(name = "ViewerDocumentService", contextualName = "create-viewer-document")
public Visualizations addLayoutGrid(Document document, boolean layerVisibilityDefaultValue, boolean visualParsingGrid) {
LayoutGrid layoutGrid = createLayoutGrid(document, visualParsingGrid);
return Visualizations.builder()
.layer(visualParsingGrid ? ContentStreams.KNECON_VISUAL_PARSING : ContentStreams.KNECON_LAYOUT)
.visualizationsOnPages(layoutGrid.getVisualizationsPerPages())
.layerVisibilityDefaultValue(layerVisibilityDefaultValue)
.build();
}
private LayoutGrid createLayoutGrid(Document document, boolean visualParsingGrid) {
LayoutGrid layoutGrid = new LayoutGrid(document.getNumberOfPages());
document.streamAllSubNodes().forEach(semanticNode -> {
Color color = switch (semanticNode.getType()) {
case PARAGRAPH -> PARAGRAPH_COLOR;
case TABLE -> TABLE_COLOR;
case SECTION -> SECTION_COLOR;
case HEADLINE -> HEADLINE_COLOR;
case HEADER, FOOTER -> HEADER_COLOR;
case IMAGE -> IMAGE_COLOR;
default -> null;
};
if (isNotSectionOrTableCellOrDocument(semanticNode)) {
addAsRectangle(semanticNode, layoutGrid, color);
}
if (semanticNode.getType().equals(NodeType.SECTION)) {
addSection(semanticNode, layoutGrid, color);
}
if (semanticNode.getType().equals(NodeType.TABLE)) {
Table table = (Table) semanticNode;
addInnerTableLines(table, layoutGrid);
}
});
document.streamAllSubNodes()
.filter(node -> (node.getEngines().contains(LayoutEngine.AI) && visualParsingGrid) || (node.getEngines().contains(LayoutEngine.ALGORITHM) && !visualParsingGrid))
.forEach(semanticNode -> {
Color color = switch (semanticNode.getType()) {
case PARAGRAPH -> PARAGRAPH_COLOR;
case TABLE -> TABLE_COLOR;
case SECTION -> SECTION_COLOR;
case HEADLINE -> HEADLINE_COLOR;
case HEADER, FOOTER -> HEADER_COLOR;
case IMAGE -> IMAGE_COLOR;
default -> null;
};
if (isNotSectionOrTableCellOrDocument(semanticNode)) {
addAsRectangle(semanticNode, layoutGrid, color);
}
if (semanticNode.getType().equals(NodeType.SECTION)) {
addSection(semanticNode, layoutGrid, color);
}
if (semanticNode.getType().equals(NodeType.TABLE)) {
Table table = (Table) semanticNode;
addInnerTableLines(table, layoutGrid);
}
});
return layoutGrid;
}
@@ -103,11 +147,11 @@ public class LayoutGridService {
List<ColoredLine> coloredLines = layoutGrid.getVisualizationsPerPages().get(page.getNumber() - 1).getColoredLines();
xs.forEach(x -> {
Line2D line = new Line2D.Double(new Point2D.Double(x, tableBBox.getMaxY()), new Point2D.Double(x, tableBBox.getMinY()));
coloredLines.add(new ColoredLine(line, INNER_LINES_COLOR));
coloredLines.add(new ColoredLine(line, INNER_LINES_COLOR, LINE_WIDTH));
});
ys.forEach(y -> {
Line2D line = new Line2D.Double(new Point2D.Double(tableBBox.getMinX(), y), new Point2D.Double(tableBBox.getMaxX(), y));
coloredLines.add(new ColoredLine(line, INNER_LINES_COLOR));
coloredLines.add(new ColoredLine(line, INNER_LINES_COLOR, LINE_WIDTH));
});
}
}
@@ -135,9 +179,9 @@ public class LayoutGridService {
List<Line2D> lines = createLinesFromRectangle(r, firstPage.getRotation());
// add string to top line
var firstLine = lines.remove(0);
coloredLines.add(new ColoredLine(firstLine, color));
coloredLines.add(new ColoredLine(firstLine, color, LINE_WIDTH));
for (Line2D line : lines) {
coloredLines.add(new ColoredLine(line, color));
coloredLines.add(new ColoredLine(line, color, LINE_WIDTH));
}
return;
}
@@ -152,16 +196,37 @@ public class LayoutGridService {
}
@SneakyThrows
private void addPlacedText(Page page, Rectangle2D textBBox, String s, LayoutGrid layoutGrid) {
Point2D.Float upperLeftCorner = switch (page.getRotation()) {
case 90 -> new Point2D.Float((float) (textBBox.getMinX()), (float) textBBox.getMinY());
case 180 -> new Point2D.Float((float) (textBBox.getMaxX()), (float) textBBox.getMinY());
case 270 -> new Point2D.Float((float) (textBBox.getMaxX()), (float) textBBox.getMaxY());
default -> new Point2D.Float((float) (textBBox.getMinX()), (float) textBBox.getMaxY());
};
// translates text, such that its right edge is a bit to the left of the drawn box
float translationAmount = ((FONT.getStringWidth(s) / 1000) * FONT_SIZE + (2 * LINE_WIDTH) + 4);
Point2D upperLeftCorner;
Point2D translationVector;
switch (page.getRotation()) {
case 90 -> {
upperLeftCorner = new Point2D.Double(textBBox.getMinX(), textBBox.getMinY());
translationVector = new Point2D.Double(FONT_SIZE, -translationAmount);
}
case 180 -> {
upperLeftCorner = new Point2D.Double(textBBox.getMaxX(), textBBox.getMinY());
translationVector = new Point2D.Double(translationAmount, FONT_SIZE);
}
case 270 -> {
upperLeftCorner = new Point2D.Double(textBBox.getMaxX(), textBBox.getMaxY());
translationVector = new Point2D.Double(-FONT_SIZE, translationAmount);
}
default -> {
upperLeftCorner = new Point2D.Double(textBBox.getMinX(), textBBox.getMaxY());
translationVector = new Point2D.Double(-translationAmount, -FONT_SIZE);
}
}
upperLeftCorner = add(upperLeftCorner, translationVector);
var placedTexts = layoutGrid.getVisualizationsPerPages().get(page.getNumber() - 1).getPlacedTexts();
placedTexts.add(new PlacedText(s, upperLeftCorner));
placedTexts.add(PlacedText.textFacingUp(s, upperLeftCorner, FONT_SIZE, Color.BLACK, FONT));
}
@@ -176,9 +241,9 @@ public class LayoutGridService {
midPageLines.remove(1);
// add string to left line
var leftLine = midPageLines.remove(1);
coloredLines.add(new ColoredLine(leftLine, color));
coloredLines.add(new ColoredLine(leftLine, color, LINE_WIDTH));
for (Line2D line : midPageLines) {
coloredLines.add(new ColoredLine(line, color));
coloredLines.add(new ColoredLine(line, color, LINE_WIDTH));
}
}
@@ -192,9 +257,9 @@ public class LayoutGridService {
lastPageLines.remove(0);
// add string to left line
var leftLine = lastPageLines.remove(2);
coloredLines.add(new ColoredLine(leftLine, color));
coloredLines.add(new ColoredLine(leftLine, color, LINE_WIDTH));
for (Line2D line : lastPageLines) {
coloredLines.add(new ColoredLine(line, color));
coloredLines.add(new ColoredLine(line, color, LINE_WIDTH));
}
}
@@ -208,9 +273,9 @@ public class LayoutGridService {
firstPageLines.remove(2);
// add string to top line
var firstLine = firstPageLines.remove(0);
coloredLines.add(new ColoredLine(firstLine, color));
coloredLines.add(new ColoredLine(firstLine, color, LINE_WIDTH));
for (Line2D line : firstPageLines) {
coloredLines.add(new ColoredLine(line, color));
coloredLines.add(new ColoredLine(line, color, LINE_WIDTH));
}
}
@@ -276,7 +341,16 @@ public class LayoutGridService {
private void addAsRectangle(SemanticNode semanticNode, LayoutGrid layoutGrid, Color color) {
semanticNode.getBBox()
.forEach((page, textBBox) -> layoutGrid.getVisualizationsPerPages().get(page.getNumber() - 1).getColoredRectangles().add(new ColoredRectangle(textBBox, color)));
.forEach((page, textBBox) -> layoutGrid.getVisualizationsPerPages()
.get(page.getNumber() - 1)
.getColoredRectangles()
.add(new ColoredRectangle(textBBox, color, LINE_WIDTH)));
}
private Point2D add(Point2D a, Point2D b) {
return new Point2D.Double(a.getX() + b.getX(), a.getY() + b.getY());
}
}
@@ -1,217 +0,0 @@
package com.knecon.fforesight.service.layoutparser.processor.services.visualization;
import java.awt.geom.AffineTransform;
import java.awt.geom.Rectangle2D;
import java.io.File;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.StandardCopyOption;
import org.apache.pdfbox.Loader;
import org.apache.pdfbox.cos.COSName;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.apache.pdfbox.pdmodel.PDDocumentCatalog;
import org.apache.pdfbox.pdmodel.PDPage;
import org.apache.pdfbox.pdmodel.PDPageContentStream;
import org.apache.pdfbox.pdmodel.PDResources;
import org.apache.pdfbox.pdmodel.font.PDFont;
import org.apache.pdfbox.pdmodel.font.PDType1Font;
import org.apache.pdfbox.pdmodel.font.Standard14Fonts;
import org.apache.pdfbox.pdmodel.graphics.optionalcontent.PDOptionalContentGroup;
import org.apache.pdfbox.pdmodel.graphics.optionalcontent.PDOptionalContentProperties;
import org.apache.pdfbox.pdmodel.graphics.state.PDExtendedGraphicsState;
import org.apache.pdfbox.util.Matrix;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.ColoredLine;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.ColoredRectangle;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.FilledRectangle;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.LayoutGrid;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.PlacedText;
import com.knecon.fforesight.service.layoutparser.processor.model.visualization.VisualizationsOnPage;
import io.micrometer.observation.Observation;
import io.micrometer.observation.ObservationRegistry;
import io.micrometer.observation.annotation.Observed;
import lombok.RequiredArgsConstructor;
import lombok.SneakyThrows;
import lombok.extern.slf4j.Slf4j;
@Slf4j
@Service
@RequiredArgsConstructor
public class ViewerDocumentService {
private static final String LAYER_NAME = "Layout grid";
private static final int FONT_SIZE = 10;
public static final float LINE_WIDTH = 1f;
private final LayoutGridService layoutGridService;
private final ObservationRegistry observationRegistry;
@SneakyThrows
@Observed(name = "ViewerDocumentService", contextualName = "create-viewer-document")
public void createViewerDocument(File originFile, Document document, File destinationFile, boolean layerVisibilityDefaultValue) {
Path tmpFile = Files.createTempFile("tmpViewerDocument", ".pdf");
PDDocument pdDocument = openPDDocument(originFile);
LayoutGrid layoutGrid = layoutGridService.createLayoutGrid(document);
PDOptionalContentGroup layer = addLayerToDocument(pdDocument, layerVisibilityDefaultValue);
PDFont font = new PDType1Font(Standard14Fonts.FontName.HELVETICA);
for (int pageNumber = 0; pageNumber < pdDocument.getNumberOfPages(); pageNumber++) {
PDPage pdPage = pdDocument.getPage(pageNumber);
//
AffineTransform textDeRotationMatrix = getTextDeRotationTransform(pdPage);
addLayerToPageResources(pdPage);
// We need to save the graphics state before, such that our appended content cannot be affected by previous content streams with side effects,
// e.g. not escaped matrix transformations.
escapePreviousContents(pdDocument, pdPage);
VisualizationsOnPage visualizationsOnPage = layoutGrid.getVisualizationsPerPages().get(pageNumber);
assert pageNumber == visualizationsOnPage.getPageNumber();
// We need to append to the content stream, otherwise the content could be overlapped by following content.
try (var contentStream = new PDPageContentStream(pdDocument, pdPage, PDPageContentStream.AppendMode.APPEND, true)) {
contentStream.beginMarkedContent(COSName.OC, layer);
contentStream.saveGraphicsState();
contentStream.setLineWidth(LINE_WIDTH);
for (ColoredLine coloredLine : visualizationsOnPage.getColoredLines()) {
contentStream.setStrokingColor(coloredLine.color());
contentStream.moveTo((float) coloredLine.line().getX1(), (float) coloredLine.line().getY1());
contentStream.lineTo((float) coloredLine.line().getX2(), (float) coloredLine.line().getY2());
contentStream.stroke();
}
for (ColoredRectangle coloredRectangle : visualizationsOnPage.getColoredRectangles()) {
contentStream.setStrokingColor(coloredRectangle.color());
Rectangle2D r = coloredRectangle.rectangle2D();
contentStream.addRect((float) r.getX(), (float) r.getY(), (float) r.getWidth(), (float) r.getHeight());
contentStream.stroke();
}
for (FilledRectangle filledRectangle : visualizationsOnPage.getFilledRectangles()) {
contentStream.setNonStrokingColor(filledRectangle.color());
PDExtendedGraphicsState graphicsState = new PDExtendedGraphicsState();
graphicsState.setNonStrokingAlphaConstant(filledRectangle.alpha());
contentStream.setGraphicsStateParameters(graphicsState);
Rectangle2D r = filledRectangle.rectangle2D();
contentStream.addRect((float) r.getX(), (float) r.getY(), (float) r.getWidth(), (float) r.getHeight());
contentStream.fill();
}
for (PlacedText placedText : visualizationsOnPage.getPlacedTexts()) {
contentStream.setFont(font, FONT_SIZE);
contentStream.beginText();
Matrix textMatrix = new Matrix((float) textDeRotationMatrix.getScaleX(),
(float) textDeRotationMatrix.getShearX(),
(float) textDeRotationMatrix.getShearY(),
(float) textDeRotationMatrix.getScaleY(),
(float) placedText.lineStart().getX(),
(float) placedText.lineStart().getY());
textMatrix.translate(-((font.getStringWidth(placedText.text()) / 1000) * FONT_SIZE + (2 * LINE_WIDTH) + 4), -FONT_SIZE);
contentStream.setTextMatrix(textMatrix);
contentStream.showText(placedText.text());
contentStream.endText();
}
contentStream.restoreGraphicsState();
contentStream.endMarkedContent();
}
if (pageNumber % 500 == 0 && pageNumber != 0) { // re-open document every once in a while to save on RAM
log.info("Incremental save after {} pages", pageNumber);
observedIncrementalSave(pdDocument, destinationFile);
pdDocument.close();
Files.copy(destinationFile.toPath(), tmpFile, StandardCopyOption.REPLACE_EXISTING);
pdDocument = openPDDocument(tmpFile.toFile());
layer = addLayerToDocument(pdDocument, layerVisibilityDefaultValue);
}
}
observedIncrementalSave(pdDocument, destinationFile);
tmpFile.toFile().delete();
pdDocument.close();
}
private static PDDocument openPDDocument(File tmpFile) throws IOException {
PDDocument pdDocument;
pdDocument = Loader.loadPDF(tmpFile);
pdDocument.setAllSecurityToBeRemoved(true);
return pdDocument;
}
@SneakyThrows
private void observedIncrementalSave(PDDocument pdDocument, File outputFile) {
Observation.createNotStarted("ViewerDocumentService", observationRegistry).contextualName("incremental-save").observe(() -> {
try {
pdDocument.save(outputFile);
} catch (IOException e) {
throw new RuntimeException(e);
}
});
}
private static void addLayerToPageResources(PDPage pdPage) {
PDResources resources = pdPage.getResources();
if (resources == null) {
resources = new PDResources();
pdPage.setResources(resources);
}
}
private static void escapePreviousContents(PDDocument pdDocument, PDPage pdPage) throws IOException {
try (var contentStream = new PDPageContentStream(pdDocument, pdPage, PDPageContentStream.AppendMode.PREPEND, false)) {
contentStream.saveGraphicsState();
}
try (var contentStream = new PDPageContentStream(pdDocument, pdPage, PDPageContentStream.AppendMode.APPEND, false)) {
contentStream.restoreGraphicsState();
}
}
private static PDOptionalContentGroup addLayerToDocument(PDDocument pdDocument, boolean layerVisibilityDefaultValue) {
PDDocumentCatalog catalog = pdDocument.getDocumentCatalog();
PDOptionalContentProperties ocprops = catalog.getOCProperties();
if (ocprops == null) {
ocprops = new PDOptionalContentProperties();
catalog.setOCProperties(ocprops);
}
PDOptionalContentGroup layer = null;
if (ocprops.hasGroup(LAYER_NAME)) {
layer = ocprops.getGroup(LAYER_NAME);
} else {
layer = new PDOptionalContentGroup(LAYER_NAME);
ocprops.addGroup(layer);
}
ocprops.setGroupEnabled(layer, layerVisibilityDefaultValue);
// dictionariesToUpdate.add(catalog.getCOSObject());
return layer;
}
private static AffineTransform getTextDeRotationTransform(PDPage page) {
return AffineTransform.getQuadrantRotateInstance(switch (page.getRotation()) {
case 90 -> 3;
case 180 -> 2;
case 270 -> 1;
default -> 0;
});
}
}
@@ -0,0 +1,28 @@
package com.knecon.fforesight.service.layoutparser.processor.utils;
import java.awt.Color;
import lombok.AccessLevel;
import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Getter;
import lombok.experimental.FieldDefaults;
@Builder
@AllArgsConstructor
@Getter
@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
public class DrawingOptions {
boolean stroke;
@Builder.Default
Color strokeColor = Color.BLACK;
@Builder.Default
float strokeWidth = 1f;
boolean fill;
@Builder.Default
Color fillColor = Color.BLACK;
}
@@ -0,0 +1,88 @@
package com.knecon.fforesight.service.layoutparser.processor.utils;
import java.awt.geom.Point2D;
import java.util.Comparator;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Cell;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Rectangle;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Ruling;
public class GeometricComparators {
private static final int COMPARATOR_ROUNDING = 2;
public static final Comparator<Point2D> X_FIRST_POINT_COMPARATOR = (point1, point2) -> {
int rv = 0;
float point1X = DoubleComparisons.round(point1.getX(), COMPARATOR_ROUNDING);
float point1Y = DoubleComparisons.round(point1.getY(), COMPARATOR_ROUNDING);
float point2X = DoubleComparisons.round(point2.getX(), COMPARATOR_ROUNDING);
float point2Y = DoubleComparisons.round(point2.getY(), COMPARATOR_ROUNDING);
if (point1X > point2X) {
rv = 1;
} else if (point1X < point2X) {
rv = -1;
} else if (point1Y > point2Y) {
rv = 1;
} else if (point1Y < point2Y) {
rv = -1;
}
return rv;
};
public static final Comparator<Point2D> Y_FIRST_POINT_COMPARATOR = (point1, point2) -> {
int rv = 0;
float point1X = DoubleComparisons.round(point1.getX(), COMPARATOR_ROUNDING);
float point1Y = DoubleComparisons.round(point1.getY(), COMPARATOR_ROUNDING);
float point2X = DoubleComparisons.round(point2.getX(), COMPARATOR_ROUNDING);
float point2Y = DoubleComparisons.round(point2.getY(), COMPARATOR_ROUNDING);
if (point1Y > point2Y) {
rv = 1;
} else if (point1Y < point2Y) {
rv = -1;
} else if (point1X > point2X) {
rv = 1;
} else if (point1X < point2X) {
rv = -1;
}
return rv;
};
public static final Comparator<Cell> CELL_SIZE_COMPARATOR = (cell1, cell2) -> {
Double cell1Size = cell1.getHeight() * cell1.getWidth();
Double cell2Size = cell2.getHeight() * cell2.getWidth();
return cell1Size.compareTo(cell2Size);
};
public static final Comparator<Rectangle> RECTANGLE_SIZE_COMPARATOR = (rect1, rect2) -> {
Double rect1Size = rect1.getHeight() * rect1.getWidth();
Double rect2Size = rect2.getHeight() * rect2.getWidth();
return rect1Size.compareTo(rect2Size);
};
public static final Comparator<Ruling> X_FIRST_RULING_COMPARATOR = (ruling1, ruling2) -> {
int rv = 0;
float point1X = DoubleComparisons.round(Math.min(ruling1.getLeft(), ruling1.getRight()), COMPARATOR_ROUNDING);
float point1Y = DoubleComparisons.round(Math.min(ruling1.getTop(), ruling1.getBottom()), COMPARATOR_ROUNDING);
float point2X = DoubleComparisons.round(Math.min(ruling2.getLeft(), ruling2.getRight()), COMPARATOR_ROUNDING);
float point2Y = DoubleComparisons.round(Math.min(ruling2.getTop(), ruling2.getBottom()), COMPARATOR_ROUNDING);
if (point1X > point2X) {
rv = 1;
} else if (point1X < point2X) {
rv = -1;
} else if (point1Y > point2Y) {
rv = 1;
} else if (point1Y < point2Y) {
rv = -1;
}
return rv;
};
}
@@ -21,11 +21,7 @@ import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Pa
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.AtomicTextBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.TextBlock;
import lombok.AccessLevel;
import lombok.Builder;
import lombok.Getter;
import lombok.SneakyThrows;
import lombok.experimental.FieldDefaults;
import lombok.experimental.UtilityClass;
import lombok.extern.slf4j.Slf4j;
@@ -41,20 +37,20 @@ public class PdfVisualisationUtility {
public void drawNode(PDDocument document, DocumentTree.Entry entry) {
Options options = buildStandardOptionsForNodes(entry);
DrawingOptions options = buildStandardOptionsForNodes(entry);
drawBBoxAndLabelAndNumberOnPage(document, entry, options);
}
public void drawTextBlock(PDDocument document, TextBlock textBlock, Options options) {
public void drawTextBlock(PDDocument document, TextBlock textBlock, DrawingOptions options) {
textBlock.getAtomicTextBlocks().forEach(atb -> drawAtomicTextBlock(document, atb, options));
}
public void drawAtomicTextBlock(PDDocument document, AtomicTextBlock atomicTextBlock, Options options) {
public void drawAtomicTextBlock(PDDocument document, AtomicTextBlock atomicTextBlock, DrawingOptions options) {
drawRectangle2DList(document, atomicTextBlock.getPage().getNumber(), atomicTextBlock.getPositions().stream().toList(), options);
@@ -62,7 +58,7 @@ public class PdfVisualisationUtility {
@SneakyThrows
public void drawText(String string, PDDocument document, Point2D location, Integer pageNumber, Options options) {
public void drawText(String string, PDDocument document, Point2D location, Integer pageNumber, DrawingOptions options) {
var pdPage = document.getPage(pageNumber - 1);
var contentStream = new PDPageContentStream(document, pdPage, PDPageContentStream.AppendMode.APPEND, true);
@@ -80,14 +76,14 @@ public class PdfVisualisationUtility {
@SneakyThrows
public void drawRectangle2DList(PDDocument document, int pageNumber, List<Rectangle2D> rectCollection, Options options) {
public void drawRectangle2DList(PDDocument document, int pageNumber, List<Rectangle2D> rectCollection, DrawingOptions options) {
var pdPage = document.getPage(pageNumber - 1);
drawRectangle2DList(document, rectCollection, options, pdPage);
}
private void drawRectangle2DList(PDDocument document, List<Rectangle2D> rectCollection, Options options, PDPage pdPage) throws IOException {
private void drawRectangle2DList(PDDocument document, List<Rectangle2D> rectCollection, DrawingOptions options, PDPage pdPage) throws IOException {
var contentStream = new PDPageContentStream(document, pdPage, PDPageContentStream.AppendMode.APPEND, true);
@@ -110,9 +106,9 @@ public class PdfVisualisationUtility {
}
private Options buildStandardOptionsForNodes(DocumentTree.Entry entry) {
private DrawingOptions buildStandardOptionsForNodes(DocumentTree.Entry entry) {
return Options.builder().stroke(true).strokeColor(switch (entry.getType()) {
return DrawingOptions.builder().stroke(true).strokeColor(switch (entry.getType()) {
case DOCUMENT -> Color.LIGHT_GRAY;
case HEADER, FOOTER -> Color.GREEN;
case PARAGRAPH -> Color.BLUE;
@@ -125,7 +121,7 @@ public class PdfVisualisationUtility {
}
private void drawBBoxAndLabelAndNumberOnPage(PDDocument document, DocumentTree.Entry entry, Options options) {
private void drawBBoxAndLabelAndNumberOnPage(PDDocument document, DocumentTree.Entry entry, DrawingOptions options) {
Map<Page, Rectangle2D> rectanglesPerPage = entry.getNode().getBBox();
rectanglesPerPage.forEach((page, rectangle2D) -> {
@@ -152,7 +148,7 @@ public class PdfVisualisationUtility {
@SneakyThrows
public static void drawLine2DList(PDDocument pdDocument, int pageNumber, List<? extends Line2D> line2DS, Options options) {
public static void drawLine2DList(PDDocument pdDocument, int pageNumber, List<? extends Line2D> line2DS, DrawingOptions options) {
var pdPage = pdDocument.getPage(pageNumber - 1);
var contentStream = new PDPageContentStream(pdDocument, pdPage, PDPageContentStream.AppendMode.APPEND, true);
@@ -176,21 +172,4 @@ public class PdfVisualisationUtility {
contentStream.close();
}
@Builder
@Getter
@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
public static class Options {
boolean fill;
boolean stroke;
@Builder.Default
Color strokeColor = Color.BLACK;
@Builder.Default
float strokeWidth = 1f;
@Builder.Default
Color fillColor = Color.BLACK;
}
}
@@ -2,6 +2,7 @@ package com.knecon.fforesight.service.layoutparser.processor.utils;
import static java.lang.String.format;
import java.awt.geom.Area;
import java.awt.geom.Rectangle2D;
import java.awt.geom.RectangularShape;
import java.util.Collections;
@@ -37,15 +38,28 @@ public class RectangleTransformations {
}
public static double calculateIntersectedArea(Rectangle2D r1, Rectangle2D r2) {
Area a1 = new Area(r1);
Area a2 = new Area(r2);
a1.intersect(a2);
Rectangle2D intersection = a1.getBounds2D();
return intersection.getWidth() * intersection.getHeight();
}
public static Rectangle2D bBoxUnionAtomicTextBlock(List<AtomicTextBlock> atomicTextBlocks) {
return atomicTextBlocks.stream().flatMap(atomicTextBlock -> atomicTextBlock.getPositions().stream()).collect(new Rectangle2DBBoxCollector());
}
public static Collector<Rectangle2D, Rectangle2DBBoxCollector.BBox, Rectangle2D> collectBBox() {
return new Rectangle2DBBoxCollector();
}
public static PDRectangle toPDRectangleBBox(List<Rectangle> rectangles) {
Rectangle2D rectangle2D = RectangleTransformations.rectangleBBox(rectangles);
@@ -70,6 +84,7 @@ public class RectangleTransformations {
return format("%f,%f,%f,%f", rectangle2D.getX(), rectangle2D.getY(), rectangle2D.getWidth(), rectangle2D.getHeight());
}
public static Rectangle2D rectangleBBox(List<Rectangle> rectangles) {
return rectangles.stream().map(RectangleTransformations::toRectangle2D).collect(new Rectangle2DBBoxCollector());
@@ -84,6 +99,7 @@ public class RectangleTransformations {
-redactionLogRectangle.getHeight());
}
public static Rectangle2D toRectangle2D(PDRectangle rectangle) {
return new Rectangle2D.Double(rectangle.getLowerLeftX(), rectangle.getLowerLeftY(), rectangle.getWidth(), rectangle.getHeight());
@@ -0,0 +1,77 @@
package com.knecon.fforesight.service.layoutparser.processor.utils;
import static com.knecon.fforesight.service.layoutparser.processor.utils.GeometricComparators.Y_FIRST_POINT_COMPARATOR;
import java.awt.geom.Point2D;
import java.awt.geom.Rectangle2D;
import java.util.ArrayList;
import java.util.List;
import java.util.Map;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Ruling;
public class RectangularIntersectionFinder {
public static List<Rectangle2D> find(List<Ruling> horizontalRulingLines, List<Ruling> verticalRulingLines) {
// Fix for 211.pdf
for (Ruling r : horizontalRulingLines) {
if (r.getX2() < r.getX1()) {
double a = r.getX2();
r.x2 = (float) r.getX1();
r.x1 = (float) a;
}
}
List<Rectangle2D> foundRectangles = new ArrayList<>();
Map<Point2D, Ruling[]> intersectionPoints = Ruling.findIntersections(horizontalRulingLines, verticalRulingLines);
List<Point2D> intersectionPointsList = new ArrayList<>(intersectionPoints.keySet());
intersectionPointsList.sort(Y_FIRST_POINT_COMPARATOR);
for (int i = 0; i < intersectionPointsList.size(); i++) {
Point2D topLeft = intersectionPointsList.get(i);
Ruling[] hv = intersectionPoints.get(topLeft);
// CrossingPointsDirectlyBelow( topLeft );
List<Point2D> xPoints = new ArrayList<>();
// CrossingPointsDirectlyToTheRight( topLeft );
List<Point2D> yPoints = new ArrayList<>();
for (Point2D p : intersectionPointsList.subList(i, intersectionPointsList.size())) {
if (p.getX() == topLeft.getX() && p.getY() > topLeft.getY()) {
xPoints.add(p);
}
if (p.getY() == topLeft.getY() && p.getX() > topLeft.getX()) {
yPoints.add(p);
}
}
outer:
for (Point2D xPoint : xPoints) {
// is there a vertical edge b/w topLeft and xPoint?
if (!hv[1].equals(intersectionPoints.get(xPoint)[1])) {
continue;
}
for (Point2D yPoint : yPoints) {
// is there a horizontal edge b/w topLeft and yPoint ?
if (!hv[0].equals(intersectionPoints.get(yPoint)[0])) {
continue;
}
Point2D btmRight = new Point2D.Float((float) yPoint.getX(), (float) xPoint.getY());
if (intersectionPoints.containsKey(btmRight)
&& intersectionPoints.get(btmRight)[0].equals(intersectionPoints.get(xPoint)[0])
&& intersectionPoints.get(btmRight)[1].equals(intersectionPoints.get(yPoint)[1])) {
foundRectangles.add(new Rectangle2D.Double(topLeft.getX(), topLeft.getY(), btmRight.getX() - topLeft.getX(), btmRight.getY() - topLeft.getY()));
break outer;
}
}
}
}
// TODO create cells for vertical ruling lines with aligned endpoints at the top/bottom of a grid
// that aren't connected with an horizontal ruler?
// see: https://github.com/jazzido/tabula-extractor/issues/78#issuecomment-41481207
return foundRectangles;
}
}
@@ -0,0 +1,172 @@
package com.knecon.fforesight.service.layoutparser.processor.utils;
import static com.knecon.fforesight.service.layoutparser.processor.utils.GeometricComparators.X_FIRST_POINT_COMPARATOR;
import static com.knecon.fforesight.service.layoutparser.processor.utils.GeometricComparators.Y_FIRST_POINT_COMPARATOR;
import java.awt.geom.Point2D;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.HashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Rectangle;
public class SpreadsheetFinder {
private static final int MAX_OUTER_POINT_TOLERANCE = 10;
private static final float AREA_TOLERANCE = 0.001f;
public static List<Rectangle> findSpreadsheetsFromCells(List<? extends Rectangle> cells) {
// via: http://stackoverflow.com/questions/13746284/merging-multiple-adjacent-rectangles-into-one-polygon
List<Rectangle> rectangles = new ArrayList<>();
Set<Point2D> pointSet = new HashSet<>();
Map<Point2D, Point2D> edgesH = new HashMap<>();
Map<Point2D, Point2D> edgesV = new HashMap<>();
for (Rectangle cell : cells) {
for (Point2D pt : cell.getPoints()) {
if (pointSet.contains(pt)) { // shared vertex, remove it
pointSet.remove(pt);
} else {
pointSet.add(pt);
}
}
}
// X first sort
List<Point2D> pointsSortX = new ArrayList<>(pointSet);
pointsSortX.sort(X_FIRST_POINT_COMPARATOR);
// Y first sort
List<Point2D> pointsSortY = new ArrayList<>(pointSet);
pointsSortY.sort(Y_FIRST_POINT_COMPARATOR);
int i = 0;
while (i < pointSet.size()) {
float currY = (float) pointsSortY.get(i).getY();
while (i < pointSet.size() && DoubleComparisons.feq(pointsSortY.get(i).getY(), currY)) {
edgesH.put(pointsSortY.get(i), pointsSortY.get(i + 1));
edgesH.put(pointsSortY.get(i + 1), pointsSortY.get(i));
i += 2;
}
}
i = 0;
while (i < pointSet.size()) {
float currX = (float) pointsSortX.get(i).getX();
while (i < pointSet.size() && DoubleComparisons.feq(pointsSortX.get(i).getX(), currX)) {
edgesV.put(pointsSortX.get(i), pointsSortX.get(i + 1));
edgesV.put(pointsSortX.get(i + 1), pointsSortX.get(i));
i += 2;
}
}
// Get all the polygons
List<List<PolygonVertex>> polygons = new ArrayList<>();
Point2D nextVertex;
while (!edgesH.isEmpty()) {
ArrayList<PolygonVertex> polygon = new ArrayList<>();
Point2D first = edgesH.keySet()
.iterator().next();
polygon.add(new PolygonVertex(first, Direction.HORIZONTAL));
edgesH.remove(first);
while (true) {
PolygonVertex curr = polygon.get(polygon.size() - 1);
PolygonVertex lastAddedVertex;
if (curr.direction == Direction.HORIZONTAL) {
nextVertex = edgesV.get(curr.point);
edgesV.remove(curr.point);
lastAddedVertex = new PolygonVertex(nextVertex, Direction.VERTICAL);
} else {
nextVertex = edgesH.get(curr.point);
edgesH.remove(curr.point);
lastAddedVertex = new PolygonVertex(nextVertex, Direction.HORIZONTAL);
}
polygon.add(lastAddedVertex);
if (lastAddedVertex.equals(polygon.get(0))) {
// closed polygon
polygon.remove(polygon.size() - 1);
break;
}
}
for (PolygonVertex vertex : polygon) {
edgesH.remove(vertex.point);
edgesV.remove(vertex.point);
}
polygons.add(polygon);
}
// calculate grid-aligned minimum area rectangles for each found polygon
for (List<PolygonVertex> poly : polygons) {
float top = Float.MAX_VALUE;
float left = Float.MAX_VALUE;
float bottom = Float.MIN_VALUE;
float right = Float.MIN_VALUE;
for (PolygonVertex pt : poly) {
top = (float) Math.min(top, pt.point.getY());
left = (float) Math.min(left, pt.point.getX());
bottom = (float) Math.max(bottom, pt.point.getY());
right = (float) Math.max(right, pt.point.getX());
}
// do not add polygons with too many outer points as they are unlikely to be tables
if (poly.size() <= MAX_OUTER_POINT_TOLERANCE) {
rectangles.add(new Rectangle(top - AREA_TOLERANCE, left - AREA_TOLERANCE, right - left + 2 * AREA_TOLERANCE, bottom - top + 2 * AREA_TOLERANCE));
}
}
return rectangles;
}
private enum Direction {
HORIZONTAL,
VERTICAL
}
static class PolygonVertex {
Point2D point;
Direction direction;
PolygonVertex(Point2D point, Direction direction) {
this.direction = direction;
this.point = point;
}
@Override
public boolean equals(Object other) {
if (this == other) {
return true;
}
if (!(other instanceof PolygonVertex)) {
return false;
}
return this.point.equals(((PolygonVertex) other).point);
}
@Override
public int hashCode() {
return this.point.hashCode();
}
@Override
public String toString() {
return String.format("%s[point=%s,direction=%s]", this.getClass().getName(), this.point.toString(), this.direction.toString());
}
}
}
@@ -30,8 +30,6 @@ public class TableMergingUtility {
if (consecutiveTable.getColCount() == originalTablePageBlock.getColCount() && !hasTableHeader(consecutiveTable) && outerBoundaryAlignsX(originalTablePageBlock,
consecutiveTable)) {
consecutiveTablesWithSameColCountAndHeaders.add(consecutiveTable);
} else {
break;
}
}
return Stream.concat(Stream.of(originalTablePageBlock), consecutiveTablesWithSameColCountAndHeaders.stream()).toList();
@@ -0,0 +1,44 @@
package com.knecon.fforesight.service.layoutparser.processor.utils;
import java.util.HashMap;
import java.util.Map;
// simple implementation of a disjoint-set data structure
// https://en.wikipedia.org/wiki/Disjoint-set_data_structure
public class UnionFind<T> {
Map<T, T> parents = new HashMap<>();
Map<T, Integer> numberOfObjects = new HashMap<>();
public T find(T node) {
if (!parents.containsKey(node)) {
parents.put(node, node);
numberOfObjects.put(node, 1);
}
if (!node.equals(parents.get(node))) {
parents.put(node, find(parents.get(node)));
}
return parents.get(node);
}
public void union(T node1, T node2) {
T root1 = find(node1);
T root2 = find(node2);
if (!root1.equals(root2)) {
if (numberOfObjects.getOrDefault(root1, 1) < numberOfObjects.getOrDefault(root2, 1)) {
parents.put(root1, root2);
numberOfObjects.put(root2, numberOfObjects.get(root2) + numberOfObjects.get(root1));
} else {
parents.put(root2, root1);
numberOfObjects.put(root1, numberOfObjects.get(root1) + numberOfObjects.get(root2));
}
}
}
}
@@ -6,7 +6,7 @@ plugins {
id("org.springframework.boot") version "3.1.5"
id("io.spring.dependency-management") version "1.1.3"
id("org.sonarqube") version "4.3.0.3225"
id("io.freefair.lombok") version "8.2.2"
id("io.freefair.lombok") version "8.4"
// id("org.graalvm.buildtools.native") version "0.9.23"
}
@@ -30,8 +30,8 @@ dependencies {
implementation(project(":layoutparser-service-internal-api"))
implementation("com.iqser.red.commons:storage-commons:2.45.0")
implementation("com.knecon.fforesight:tenant-commons:0.19.0")
implementation("com.knecon.fforesight:tracing-commons:0.3.0")
implementation("com.knecon.fforesight:tenant-commons:0.21.0")
implementation("com.knecon.fforesight:tracing-commons:0.5.0")
implementation("org.springframework.boot:spring-boot-starter-actuator:${springBootStarterVersion}")
implementation("org.springframework.boot:spring-boot-starter-amqp:${springBootStarterVersion}")
implementation("com.amazonaws:aws-java-sdk-s3:1.12.536")
@@ -39,6 +39,7 @@ dependencies {
implementation("net.logstash.logback:logstash-logback-encoder:7.4")
// for integration testing only
testImplementation(project(":viewer-doc-processor"))
testImplementation(project(":layoutparser-service-internal-api"))
testImplementation("org.springframework.boot:spring-boot-starter-amqp:${springBootStarterVersion}")
@@ -27,8 +27,10 @@ import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Do
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.TextBlock;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.image.ImageServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.services.factory.DocumentGraphFactory;
import com.knecon.fforesight.service.layoutparser.processor.services.mapper.TaasDocumentDataMapper;
import com.knecon.fforesight.service.layoutparser.processor.utils.DrawingOptions;
import com.knecon.fforesight.service.layoutparser.server.utils.AbstractTest;
import com.knecon.fforesight.service.layoutparser.server.utils.visualizations.PdfDraw;
@@ -50,6 +52,7 @@ public class BdrJsonBuildTest extends AbstractTest {
file,
new ImageServiceResponse(),
new TableServiceResponse(),
new VisualLayoutParsingResponse(),
file.toString()));
}
@@ -111,7 +114,7 @@ public class BdrJsonBuildTest extends AbstractTest {
try (PDDocument pdDocument = Loader.loadPDF(file); var outputStream = new FileOutputStream(resultingFileName)) {
PdfDraw.drawDocumentGraph(pdDocument, document);
PdfDraw.drawTextBlock(pdDocument, textBlock, PdfDraw.Options.builder().stroke(true).strokeWidth(0.1f).strokeColor(Color.YELLOW).build());
PdfDraw.drawTextBlock(pdDocument, textBlock, DrawingOptions.builder().stroke(true).strokeWidth(0.1f).strokeColor(Color.YELLOW).build());
pdDocument.save(outputStream);
}
}
@@ -33,6 +33,7 @@ import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Do
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.SemanticNode;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.image.ImageServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.services.factory.DocumentGraphFactory;
import com.knecon.fforesight.tenantcommons.TenantsClient;
@@ -98,6 +99,7 @@ public class HeadlinesGoldStandardIntegrationTest {
pdfFileResource.getFile(),
new ImageServiceResponse(),
new TableServiceResponse(),
new VisualLayoutParsingResponse(),
filePath));
var foundHeadlines = documentGraph.streamAllSubNodes()
@@ -148,6 +150,7 @@ public class HeadlinesGoldStandardIntegrationTest {
@EqualsAndHashCode
@AllArgsConstructor
@ToString
@SuppressWarnings("PMD.AvoidFieldNameMatchingTypeName")
private class Headline {
private int page;
@@ -16,6 +16,7 @@ import com.knecon.fforesight.service.layoutparser.internal.api.queue.LayoutParsi
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.image.ImageServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.services.factory.DocumentGraphFactory;
import com.knecon.fforesight.service.layoutparser.processor.services.mapper.DocumentDataMapper;
import com.knecon.fforesight.service.layoutparser.server.utils.BuildDocumentTest;
@@ -58,6 +59,7 @@ public class DocumentGraphJsonWritingTest extends BuildDocumentTest {
filename.toFile(),
new ImageServiceResponse(),
new TableServiceResponse(),
new VisualLayoutParsingResponse(),
filename.toFile().toString()));
DocumentData documentData = DocumentDataMapper.toDocumentData(documentGraph);
@@ -13,6 +13,7 @@ import org.springframework.core.io.ClassPathResource;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.TextBlock;
import com.knecon.fforesight.service.layoutparser.processor.utils.DrawingOptions;
import com.knecon.fforesight.service.layoutparser.server.utils.BuildDocumentTest;
import com.knecon.fforesight.service.layoutparser.server.utils.visualizations.PdfDraw;
@@ -70,7 +71,7 @@ public class DocumentGraphVisualizationTest extends BuildDocumentTest {
try (PDDocument pdDocument = Loader.loadPDF(fileResource.getFile())) {
log.info("drawing document");
PdfDraw.drawDocumentGraph(pdDocument, documentGraph);
PdfDraw.drawTextBlock(pdDocument, textBlock, PdfDraw.Options.builder().stroke(true).strokeWidth(0.1f).strokeColor(Color.YELLOW).build());
PdfDraw.drawTextBlock(pdDocument, textBlock, DrawingOptions.builder().stroke(true).strokeWidth(0.1f).strokeColor(Color.YELLOW).build());
log.info("saving document");
pdDocument.save(tmpFile);
log.info("saved document");
@@ -3,14 +3,20 @@ package com.knecon.fforesight.service.layoutparser.server.graph;
import java.io.File;
import java.nio.file.Path;
import org.junit.jupiter.api.Disabled;
import org.junit.jupiter.api.Test;
import org.springframework.core.io.ClassPathResource;
import com.iqser.red.commons.jackson.ObjectMapperFactory;
import com.knecon.fforesight.service.layoutparser.internal.api.queue.LayoutParsingType;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.image.ImageServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.services.factory.DocumentGraphFactory;
import com.knecon.fforesight.service.layoutparser.processor.services.visualization.LayoutGridService;
import com.knecon.fforesight.service.layoutparser.processor.services.visualization.ViewerDocumentService;
import com.knecon.fforesight.service.layoutparser.server.utils.BuildDocumentTest;
import com.knecon.fforesight.service.viewerdoc.service.ViewerDocumentService;
import lombok.SneakyThrows;
@@ -20,13 +26,44 @@ public class ViewerDocumentTest extends BuildDocumentTest {
@SneakyThrows
public void testViewerDocument() {
String fileName = "files/S-Metolachlor_RAR_01_Volume_1_2018-09-06.pdf";
String fileName = "files/SinglePages/T5 VV-640252-Page16.pdf";
String tmpFileName = "/tmp/" + Path.of(fileName).getFileName() + "_VIEWER.pdf";
LayoutGridService layoutGridService = new LayoutGridService();
ViewerDocumentService viewerDocumentService = new ViewerDocumentService(layoutGridService, null);
Document document = buildGraph(fileName, LayoutParsingType.REDACT_MANAGER);
var documentFile = new ClassPathResource(fileName).getFile();
viewerDocumentService.createViewerDocument(documentFile, document, new File(tmpFileName), true);
ViewerDocumentService viewerDocumentService = new ViewerDocumentService(null);
LayoutGridService layoutGridService = new LayoutGridService(viewerDocumentService);
Document document = buildGraph(fileName, LayoutParsingType.REDACT_MANAGER);
long start = System.currentTimeMillis();
layoutGridService.addLayoutGrid(documentFile, document, new File(tmpFileName), true);
System.out.printf("Total time: %.2fs%n", ((float) (System.currentTimeMillis() - start)) / 1000);
}
@Test
@Disabled
@SneakyThrows
public void testViewerDocumentWithTables() {
String fileName = "files/cv_tables/brokenTablesOnOcr_ocred.pdf";
String tableFileName = "files/cv_tables/brokenTablesOnOcr_ocred.TABLES.json";
String tmpFileName = "/tmp/" + Path.of(fileName).getFileName() + "_VIEWER.pdf";
var mapper = ObjectMapperFactory.create();
var tableResponse = mapper.readValue(new ClassPathResource(tableFileName).getInputStream(), TableServiceResponse.class);
var documentFile = new ClassPathResource(fileName).getFile();
var classificationDocument = layoutParsingPipeline.parseLayout(LayoutParsingType.DOCUMINE,
documentFile,
new ImageServiceResponse(),
tableResponse,
new VisualLayoutParsingResponse(),Path.of(fileName).getFileName().toFile().toString());
ViewerDocumentService viewerDocumentService = new ViewerDocumentService(null);
LayoutGridService layoutGridService = new LayoutGridService(viewerDocumentService);
Document document = DocumentGraphFactory.buildDocumentGraph(classificationDocument);
layoutGridService.addLayoutGrid(documentFile, document, new File(tmpFileName), true);
}
}
@@ -29,10 +29,9 @@ import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Im
import com.knecon.fforesight.service.layoutparser.processor.model.image.ClassifiedImage;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Cell;
import com.knecon.fforesight.service.layoutparser.processor.model.table.TablePageBlock;
import com.knecon.fforesight.service.layoutparser.processor.python_api.adapter.CvTableParsingAdapter;
import com.knecon.fforesight.service.layoutparser.processor.python_api.adapter.ImageServiceResponseAdapter;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.image.ImageServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.services.SectionsBuilderService;
import com.knecon.fforesight.service.layoutparser.processor.services.classification.RedactManagerClassificationService;
import com.knecon.fforesight.service.layoutparser.server.utils.AbstractTest;
@@ -50,12 +49,6 @@ public class PdfSegmentationServiceTest extends AbstractTest {
@Autowired
private RedactManagerClassificationService redactManagerClassificationService;
@Autowired
private CvTableParsingAdapter cvTableParsingAdapter;
@Autowired
private ImageServiceResponseAdapter imageServiceResponseAdapter;
@Autowired
private SectionsBuilderService sectionsBuilderService;
@@ -64,10 +57,11 @@ public class PdfSegmentationServiceTest extends AbstractTest {
public ClassificationDocument buildClassificationDocument(File originDocument, TableServiceResponse tableServiceResponse) {
ClassificationDocument classificationDocument = layoutParsingPipeline.parseLayout(LayoutParsingType.REDACT_MANAGER,
originDocument,
new ImageServiceResponse(),
tableServiceResponse,
"document");
originDocument,
new ImageServiceResponse(),
tableServiceResponse,
new VisualLayoutParsingResponse(),
"document");
redactManagerClassificationService.classifyDocument(classificationDocument);
@@ -87,11 +81,11 @@ public class PdfSegmentationServiceTest extends AbstractTest {
@Test
public void tablesToHtmlDebugger() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/A20622A izRMS (CZ) fRR Part B9_Page185.pdf");
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/T5 VV-640252-Page16.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
toHtml(document, "/tmp/A20622A izRMS (CZ) fRR Part B9_Page185.html");
toHtml(document, "/tmp/T5.html");
}
@@ -109,6 +103,7 @@ public class PdfSegmentationServiceTest extends AbstractTest {
}
@Disabled
@Test
public void testScanRotationBorderIsIgnored() throws IOException {
@@ -117,8 +112,16 @@ public class PdfSegmentationServiceTest extends AbstractTest {
var tableServiceResponse = objectMapper.readValue(cvTablesResource.getInputStream(), TableServiceResponse.class);
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile(), tableServiceResponse);
assertThat(document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList())).isNotEmpty();
var tables = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).toList();
assertThat(document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.collect(Collectors.toList())).isNotEmpty();
var tables = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList();
// Quality of the table parsing is not good, because the file is rotated at scanning.
// We only asset that the table border is not the page border.
@@ -140,12 +143,12 @@ public class PdfSegmentationServiceTest extends AbstractTest {
imageServiceResponse.getData()
.forEach(imageMetadata -> images.computeIfAbsent(imageMetadata.getPosition().getPageNumber(), x -> new ArrayList<>())
.add(new ClassifiedImage(new Rectangle2D.Double(imageMetadata.getPosition().getX1(),
imageMetadata.getPosition().getY1(),
imageMetadata.getGeometry().getWidth(),
imageMetadata.getGeometry().getHeight()),
ImageType.valueOf(imageMetadata.getClassification().getLabel().toUpperCase(Locale.ROOT)),
imageMetadata.isAlpha(),
imageMetadata.getPosition().getPageNumber())));
imageMetadata.getPosition().getY1(),
imageMetadata.getGeometry().getWidth(),
imageMetadata.getGeometry().getHeight()),
ImageType.valueOf(imageMetadata.getClassification().getLabel().toUpperCase(Locale.ROOT)),
imageMetadata.isAlpha(),
imageMetadata.getPosition().getPageNumber())));
System.out.println("object");
}
@@ -157,11 +160,22 @@ public class PdfSegmentationServiceTest extends AbstractTest {
ClassPathResource pdfFileResource = new ClassPathResource("files/Minimal Examples/Spanning Cells.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
assertThat(document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList())).isNotEmpty();
TablePageBlock table = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList()).get(0);
assertThat(document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.collect(Collectors.toList())).isNotEmpty();
TablePageBlock table = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(0);
assertThat(table.getColCount()).isEqualTo(6);
assertThat(table.getRowCount()).isEqualTo(13);
assertThat(table.getRows().stream().mapToInt(List::size).sum()).isEqualTo(6 * 13);
assertThat(table.getRows()
.stream()
.mapToInt(List::size).sum()).isEqualTo(6 * 13);
}
@@ -171,15 +185,37 @@ public class PdfSegmentationServiceTest extends AbstractTest {
ClassPathResource pdfFileResource = new ClassPathResource("files/Minimal Examples/Merge Table.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
assertThat(document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList())).isNotEmpty();
TablePageBlock firstTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList()).get(0);
assertThat(document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.collect(Collectors.toList())).isNotEmpty();
TablePageBlock firstTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(0);
assertThat(firstTable.getColCount()).isEqualTo(8);
assertThat(firstTable.getRowCount()).isEqualTo(1);
TablePageBlock secondTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList()).get(1);
TablePageBlock secondTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(1);
assertThat(secondTable.getColCount()).isEqualTo(8);
assertThat(secondTable.getRowCount()).isEqualTo(2);
List<List<Cell>> firstTableHeaderCells = firstTable.getRows().get(0).stream().map(Collections::singletonList).collect(Collectors.toList());
assertThat(secondTable.getRows().stream().allMatch(row -> row.stream().map(Cell::getHeaderCells).toList().equals(firstTableHeaderCells))).isTrue();
List<List<Cell>> firstTableHeaderCells = firstTable.getRows()
.get(0)
.stream()
.map(Collections::singletonList)
.collect(Collectors.toList());
assertThat(secondTable.getRows()
.stream()
.allMatch(row -> row.stream()
.map(Cell::getHeaderCells)
.toList().equals(firstTableHeaderCells))).isTrue();
}
@@ -189,15 +225,37 @@ public class PdfSegmentationServiceTest extends AbstractTest {
ClassPathResource pdfFileResource = new ClassPathResource("files/Minimal Examples/Merge Multi Page Table.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
assertThat(document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList())).isNotEmpty();
TablePageBlock firstTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList()).get(0);
assertThat(document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.collect(Collectors.toList())).isNotEmpty();
TablePageBlock firstTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(0);
assertThat(firstTable.getColCount()).isEqualTo(9);
assertThat(firstTable.getRowCount()).isEqualTo(5);
TablePageBlock secondTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList()).get(1);
TablePageBlock secondTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(1);
assertThat(secondTable.getColCount()).isEqualTo(9);
assertThat(secondTable.getRowCount()).isEqualTo(6);
List<List<Cell>> firstTableHeaderCells = firstTable.getRows().get(firstTable.getRowCount() - 1).stream().map(Cell::getHeaderCells).collect(Collectors.toList());
assertThat(secondTable.getRows().stream().allMatch(row -> row.stream().map(Cell::getHeaderCells).toList().equals(firstTableHeaderCells))).isTrue();
List<List<Cell>> firstTableHeaderCells = firstTable.getRows()
.get(firstTable.getRowCount() - 1)
.stream()
.map(Cell::getHeaderCells)
.collect(Collectors.toList());
assertThat(secondTable.getRows()
.stream()
.allMatch(row -> row.stream()
.map(Cell::getHeaderCells)
.toList().equals(firstTableHeaderCells))).isTrue();
}
@@ -207,19 +265,41 @@ public class PdfSegmentationServiceTest extends AbstractTest {
ClassPathResource pdfFileResource = new ClassPathResource("files/Minimal Examples/Rotated Table Headers.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
assertThat(document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList())).isNotEmpty();
TablePageBlock firstTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList()).get(0);
assertThat(document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.collect(Collectors.toList())).isNotEmpty();
TablePageBlock firstTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(0);
assertThat(firstTable.getColCount()).isEqualTo(8);
assertThat(firstTable.getRowCount()).isEqualTo(1);
TablePageBlock secondTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.toList()).get(1);
TablePageBlock secondTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(1);
assertThat(secondTable.getColCount()).isEqualTo(8);
assertThat(secondTable.getRowCount()).isEqualTo(6);
List<List<Cell>> firstTableHeaderCells = firstTable.getRows().get(0).stream().map(Collections::singletonList).collect(Collectors.toList());
assertThat(secondTable.getRows().stream().allMatch(row -> row.stream().map(Cell::getHeaderCells).toList().equals(firstTableHeaderCells))).isTrue();
List<List<Cell>> firstTableHeaderCells = firstTable.getRows()
.get(0)
.stream()
.map(Collections::singletonList)
.collect(Collectors.toList());
assertThat(secondTable.getRows()
.stream()
.allMatch(row -> row.stream()
.map(Cell::getHeaderCells)
.toList().equals(firstTableHeaderCells))).isTrue();
}
@Test // Non-sense test
@Test
public void testDoc56Page170() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/56 Fludioxonil_RAR_12_Volume_3CA_B-7_2018-02-21_Page170.pdf");
@@ -230,8 +310,8 @@ public class PdfSegmentationServiceTest extends AbstractTest {
validateTable(document, 0, 1, 1, 0, 0);
validateTable(document, 1, 2, 2, 0, 0);
validateTable(document, 2, 6, 20, 0, 0);
validateTable(document, 3, 7, 31, 0, 0);
validateTable(document, 2, 4, 19, 12, 0);
validateTable(document, 3, 2, 12, 0, 0);
}
@@ -265,29 +345,30 @@ public class PdfSegmentationServiceTest extends AbstractTest {
validateTable(document, 0, 8, 8, 0, 0);
List<List<String>> values = Arrays.asList(Arrays.asList("Annex point Reference within DAR/RAR",
"Author, date",
"Study title",
"Analytical method Author, date, No.",
"Technique, LOQ of the method, validated working range",
"Method meets analytical validation criteria",
"Remarks (in case validation criteria are not met)",
"Acceptability of the method"),
Arrays.asList("Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies"),
Arrays.asList("CA 7.1.2.1.1 DAR (2009)",
"Evans P.G. 2001 TMJ4569B, VV-323245",
"Azoxystrobin Laboratory Degradation Study in Three Soil Types, Sampled from Holland and the United Kingdom",
"Method: RAM 269 Johnson R.I., Tummon O.J., Earl M. 1995 RJ1864B, VV-377731 Johnson R.I., Tummon O.J., Earl M. 1998 RAM 269/02, VV-124072 Johnson R.I., Tummon O.J., Earl M. 2000 RAM 269/03, VV-123986 Validation: Robinson N.J. 2001 TMJ4617B, VV-895845",
"LC-MS/MS LOQ: 0.01 mg/kg (R401553 (SYN50165 7), R402173 (SYN501114 )) or 0.02 mg/kg (azoxystrobin, R230310, R234886) Working range: 0.02-1.0 or 0.01-0.5 mg/kg (depending on analyte) Other supporting quantificati on methods: HPLC-UV GC-MSD",
"Y",
"N/A",
"Y"));
"Author, date",
"Study title",
"Analytical method Author, date, No.",
"Technique, LOQ of the method, validated working range",
"Method meets analytical validation criteria",
"Remarks (in case validation criteria are not met)",
"Acceptability of the method"),
Arrays.asList(
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies",
"Part (a) Methods in soil, water, sediment, air and any additional matrices used in support of environmental fate studies"),
Arrays.asList("CA 7.1.2.1.1 DAR (2009)",
"Evans P.G. 2001 TMJ4569B, VV-323245",
"Azoxystrobin Laboratory Degradation Study in Three Soil Types, Sampled from Holland and the United Kingdom",
"Method: RAM 269 Johnson R.I., Tummon O.J., Earl M. 1995 RJ1864B, VV-377731 Johnson R.I., Tummon O.J., Earl M. 1998 RAM 269/02, VV-124072 Johnson R.I., Tummon O.J., Earl M. 2000 RAM 269/03, VV-123986 Validation: Robinson N.J. 2001 TMJ4617B, VV-895845",
"LC-MS/MS LOQ: 0.01 mg/kg (R401553 (SYN50165 7), R402173 (SYN501114 )) or 0.02 mg/kg (azoxystrobin, R230310, R234886) Working range: 0.02-1.0 or 0.01-0.5 mg/kg (depending on analyte) Other supporting quantificati on methods: HPLC-UV GC-MSD",
"Y",
"N/A",
"Y"));
validateTable(document, 0, values);
@@ -579,10 +660,108 @@ public class PdfSegmentationServiceTest extends AbstractTest {
}
@Test
public void testT0() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/T0 TableWithMergedCells.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
validateTableSize(document, 1);
validateTable(document, 0, 6, 8, 0, 0);
}
@Test
public void testT1() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/T1 MultipleNestedTable.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
validateTableSize(document, 4);
validateTable(document, 0, 3, 3, 0, 0);
validateTable(document, 1, 3, 6, 2, 0);
validateTable(document, 2, 3, 3, 1, 0);
validateTable(document, 3, 3, 3, 0, 0);
}
@Test
public void testT2() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/T2 MultipleTables.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
validateTableSize(document, 6);
validateTable(document, 0, 5, 5, 0, 0);
validateTable(document, 1, 5, 6, 0, 0);
validateTable(document, 2, 5, 5, 0, 0);
validateTable(document, 3, 5, 5, 0, 0);
validateTable(document, 4, 5, 5, 0, 0);
validateTable(document, 5, 5, 5, 0, 0);
}
@Test
public void testT3() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/T3 S-Meto_Page29.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
validateTableSize(document, 1);
validateTable(document, 0, 6, 5, 0, 0);
}
@Test
public void testT4() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/T4 138 IDD0000261736_Page16.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
validateTableSize(document, 1);
validateTable(document, 0, 5, 8, 1, 0);
}
@Test
public void testT5() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/T5 VV-640252-Page16.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
validateTableSize(document, 5);
validateTable(document, 0, 1, 1, 0, 0);
validateTable(document, 1, 1, 1, 0, 0);
validateTable(document, 2, 1, 1, 0, 0);
validateTable(document, 3, 1, 1, 0, 0);
validateTable(document, 4, 1, 1, 0, 0);
}
@SneakyThrows
private void toHtml(ClassificationDocument document, String filename) {
var tables = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).toList();
var tables = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList();
StringBuilder sb = new StringBuilder();
int currentPage = 1;
@@ -603,9 +782,19 @@ public class PdfSegmentationServiceTest extends AbstractTest {
private void validateTable(ClassificationDocument document, int tableIndex, int colCount, int rowCount, int emptyCellsCountCorrect, int emptyCellsCountIncorrect) {
TablePageBlock table = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).toList().get(tableIndex);
TablePageBlock table = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(tableIndex);
List<List<Cell>> rows = table.getRows();
int emptyCellsFoundFound = rows.stream().flatMap(List::stream).toList().stream().filter(f -> f.toString().equals("")).toList().size();
int emptyCellsFoundFound = rows.stream()
.flatMap(List::stream)
.toList()
.stream()
.filter(f -> f.toString().isEmpty())
.toList().size();
for (List<Cell> row : table.getRows()) {
row.forEach(r -> System.out.println(r.toString()));
@@ -620,11 +809,20 @@ public class PdfSegmentationServiceTest extends AbstractTest {
private void validateTable(ClassificationDocument document, int tableIndex, List<List<String>> values) {
TablePageBlock table = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).toList().get(tableIndex);
TablePageBlock table = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(tableIndex);
List<List<Cell>> rows = table.getRows();
List<Cell> rowsFlattened = rows.stream().flatMap(List::stream).toList();
List<String> valuesFlattened = values.stream().flatMap(List::stream).toList();
List<Cell> rowsFlattened = rows.stream()
.flatMap(List::stream)
.toList();
List<String> valuesFlattened = values.stream()
.flatMap(List::stream)
.toList();
for (int i = 0; i < valuesFlattened.size(); i++) {
Cell cell = rowsFlattened.get(i);
@@ -637,7 +835,11 @@ public class PdfSegmentationServiceTest extends AbstractTest {
private void validateTableSize(ClassificationDocument document, int tableSize) {
assertThat(document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).toList().size()).isEqualTo(tableSize);
assertThat(document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList().size()).isEqualTo(tableSize);
}
@@ -1,13 +1,17 @@
package com.knecon.fforesight.service.layoutparser.server.services;
import java.awt.Color;
import java.awt.geom.Rectangle2D;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.Collections;
import java.util.LinkedList;
import java.util.List;
import java.util.stream.Collectors;
import org.apache.pdfbox.Loader;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.junit.jupiter.api.Disabled;
import org.junit.jupiter.api.Test;
import org.springframework.core.io.ClassPathResource;
@@ -21,13 +25,14 @@ import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Ta
import com.knecon.fforesight.service.layoutparser.processor.model.table.CleanRulings;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.image.ImageServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.services.PageContentExtractor;
import com.knecon.fforesight.service.layoutparser.processor.services.RulingCleaningService;
import com.knecon.fforesight.service.layoutparser.processor.services.factory.DocumentGraphFactory;
import com.knecon.fforesight.service.layoutparser.processor.services.mapper.DocumentDataMapper;
import com.knecon.fforesight.service.layoutparser.processor.services.mapper.PropertiesMapper;
import com.knecon.fforesight.service.layoutparser.processor.services.visualization.LayoutGridService;
import com.knecon.fforesight.service.layoutparser.processor.services.visualization.ViewerDocumentService;
import com.knecon.fforesight.service.layoutparser.processor.utils.DrawingOptions;
import com.knecon.fforesight.service.layoutparser.processor.utils.RectangularIntersectionFinder;
import com.knecon.fforesight.service.layoutparser.server.utils.BuildDocumentTest;
import com.knecon.fforesight.service.layoutparser.server.utils.visualizations.PdfDraw;
@@ -36,19 +41,40 @@ import lombok.SneakyThrows;
public class RulingCleaningServiceTest extends BuildDocumentTest {
@Test
// @Disabled
@Disabled
@SneakyThrows
public void textRectanglesFromRulingsExtraction() {
String fileName = "files/SinglePages/T5 VV-640252-Page16.pdf";
String lineFileName = "/tmp/" + Path.of(fileName).getFileName().toString() + "_CELLS.pdf";
List<PageContents> pageContents = PageContentExtractor.getSortedPageContents(fileName);
RulingCleaningService rulingCleaningService = new RulingCleaningService();
List<List<Rectangle2D>> rectanglesPerPage = new LinkedList<>();
for (PageContents pageContent : pageContents) {
CleanRulings cleanRulings = rulingCleaningService.getCleanRulings(Collections.emptyList(), pageContent.getRulings());
List<Rectangle2D> rects = RectangularIntersectionFinder.find(cleanRulings.getHorizontal(), cleanRulings.getVertical());
rectanglesPerPage.add(rects);
}
PdfDraw.drawRectanglesPerPage(fileName, rectanglesPerPage, lineFileName, DrawingOptions.builder().stroke(true).strokeColor(Color.RED).build());
}
@Test
@Disabled
@SneakyThrows
public void textRulingExtraction() {
String fileName = "files/211.pdf";
String fileName = "files/SinglePages/19 Chlorothalonil RAR 08 Volume 3CA B 6b metabolites Oct 2017_Page35.pdf";
String lineFileName = "/tmp/" + Path.of(fileName).getFileName().toString() + "_LINES.pdf";
List<PageContents> pageContents = PageContentExtractor.getSortedPageContents(fileName);
RulingCleaningService rulingCleaningService = new RulingCleaningService();
PdfDraw.drawLinesPerPage(fileName, pageContents.stream().map(PageContents::getRulings).toList(), lineFileName);
List<CleanRulings> cleanRulingsPerPage = new LinkedList<>();
for (PageContents pageContent : pageContents) {
cleanRulingsPerPage.add(rulingCleaningService.getCleanRulings(Collections.emptyList(), pageContent.getRulings()));
}
var cleanRulings = cleanRulingsPerPage.stream().map(CleanRulings::getVertical).collect(Collectors.toList());
PdfDraw.drawLinesPerPage(fileName, cleanRulings, lineFileName);
}
@@ -57,9 +83,6 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
@SneakyThrows
public void testTableExtraction() {
LayoutGridService layoutGridService = new LayoutGridService();
ViewerDocumentService viewerDocumentService = new ViewerDocumentService(layoutGridService, null);
ClassPathResource resource = new ClassPathResource("files");
List<String> pdfFileNames = Files.walk(resource.getFile().toPath())
.filter(path -> path.getFileName().toString().endsWith(".pdf"))
@@ -67,8 +90,8 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
.map(Path::toString)
.toList();
for (int i = 0; i < pdfFileNames.size(); i++) {
writeJsons(Path.of(pdfFileNames.get(i)));
for (String pdfFileName : pdfFileNames) {
writeJsons(Path.of(pdfFileName));
}
}
@@ -80,21 +103,23 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
filename.toFile(),
new ImageServiceResponse(),
new TableServiceResponse(),
new VisualLayoutParsingResponse(),
filename.toFile().toString()));
Document documentGraphAfter = DocumentGraphFactory.buildDocumentGraph(layoutParsingPipeline.parseLayout(LayoutParsingType.REDACT_MANAGER,
filename.toFile(),
new ImageServiceResponse(),
new TableServiceResponse(),
new VisualLayoutParsingResponse(),
filename.toFile().toString()));
DocumentData documentDataBefore = DocumentDataMapper.toDocumentData(documentGraphBefore);
DocumentData documentDataAfter = DocumentDataMapper.toDocumentData(documentGraphAfter);
if (!compareStructures(documentDataBefore.getDocumentStructure(), documentDataAfter.getDocumentStructure(), filename.getFileName().toString())) {
String tmpFileNameBefore = "C:/Users/YANNIK~1/AppData/Local/Temp/before." + filename.getFileName().toString();
if (!compareStructures(documentDataBefore.getDocumentStructure(), documentDataAfter.getDocumentStructure())) {
String tmpFileNameBefore = "/tmp/before." + filename.getFileName().toString();
try (PDDocument pdDocument = Loader.loadPDF(filename.toFile())) {
PdfDraw.drawDocumentGraph(pdDocument, documentGraphBefore);
pdDocument.save(tmpFileNameBefore);
}
String tmpFileNameAfter = "C:/Users/YANNIK~1/AppData/Local/Temp/after." + filename.getFileName().toString();
String tmpFileNameAfter = "/tmp/after." + filename.getFileName().toString();
try (PDDocument pdDocument = Loader.loadPDF(filename.toFile())) {
PdfDraw.drawDocumentGraph(pdDocument, documentGraphAfter);
pdDocument.save(tmpFileNameAfter);
@@ -105,9 +130,9 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
@SneakyThrows
private boolean compareStructures(DocumentStructure structure1, DocumentStructure structure2, String pdfName) {
private boolean compareStructures(DocumentStructure structure1, DocumentStructure structure2) {
List listStructure1 = structure1.streamAllEntries()
List<Table> listStructure1 = structure1.streamAllEntries()
.filter(entryData -> entryData.getType().equals(NodeType.TABLE))
.map(DocumentStructure.EntryData::getProperties)
.map(properties -> {
@@ -117,7 +142,7 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
})
.toList();
List listStructure2 = structure2.streamAllEntries()
List<Table> listStructure2 = structure2.streamAllEntries()
.filter(entryData -> entryData.getType().equals(NodeType.TABLE))
.map(DocumentStructure.EntryData::getProperties)
.map(properties -> {
@@ -128,8 +153,8 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
.toList();
for (int i = 0; i < listStructure1.size(); i++) {
Table tableNode1 = (Table) listStructure1.get(i);
Table tableNode2 = (Table) listStructure2.get(i);
Table tableNode1 = listStructure1.get(i);
Table tableNode2 = listStructure2.get(i);
if (tableNode1.getNumberOfRows() != tableNode2.getNumberOfRows() || tableNode1.getNumberOfCols() != tableNode2.getNumberOfCols()) {
return false;
}
@@ -20,6 +20,7 @@ import org.springframework.context.annotation.Import;
import org.springframework.context.annotation.Primary;
import org.springframework.core.io.ClassPathResource;
import org.springframework.test.context.junit.jupiter.SpringExtension;
import org.xmlunit.builder.Input;
import com.iqser.red.commons.jackson.ObjectMapperFactory;
import com.iqser.red.storage.commons.service.StorageService;
@@ -51,6 +52,8 @@ public abstract class AbstractTest {
private RabbitTemplate rabbitTemplate;
protected final static String ORIGIN_FILE_ID = "origin";
protected final static String VISUAL_LAYOUT_FILE = "visual";
protected final static String TABLE_FILE_ID = "table";
protected final static String IMAGE_FILE_ID = "image";
protected final static String STRUCTURE_FILE_ID = "structure";
@@ -96,7 +99,7 @@ public abstract class AbstractTest {
@SneakyThrows
protected LayoutParsingRequest prepareStorage(String file) {
return prepareStorage(file, "cv_table_parsing_response/empty.json", "image_service_response/empty.json");
return prepareStorage(file, "cv_table_parsing_response/empty.json", "image_service_response/empty.json","visual_layout_parsing_response/empty.json");
}
@@ -116,6 +119,7 @@ public abstract class AbstractTest {
.originFileStorageId(ORIGIN_FILE_ID)
.tablesFileStorageId(Optional.of(TABLE_FILE_ID))
.imagesFileStorageId(Optional.of(IMAGE_FILE_ID))
.visualLayoutParsingFileId(Optional.of(VISUAL_LAYOUT_FILE))
.structureFileStorageId(STRUCTURE_FILE_ID)
.textBlockFileStorageId(TEXT_FILE_ID)
.positionBlockFileStorageId(POSITION_FILE_ID)
@@ -136,6 +140,16 @@ public abstract class AbstractTest {
return prepareStorage(pdfFileResource.getInputStream(), cvServiceResponseFileResource.getInputStream(), imageInfoFileResource.getInputStream());
}
@SneakyThrows
protected LayoutParsingRequest prepareStorage(String file, String cvServiceResponseFile, String imageInfoFile, String visualLayoutParsingResponseFile) {
ClassPathResource pdfFileResource = new ClassPathResource(file);
ClassPathResource cvServiceResponseFileResource = new ClassPathResource(cvServiceResponseFile);
ClassPathResource imageInfoFileResource = new ClassPathResource(imageInfoFile);
ClassPathResource visualLayoutParsingResponseResource = new ClassPathResource(visualLayoutParsingResponseFile);
return prepareStorage(pdfFileResource.getInputStream(), cvServiceResponseFileResource.getInputStream(), imageInfoFileResource.getInputStream(), visualLayoutParsingResponseResource.getInputStream());
}
@SneakyThrows
protected LayoutParsingRequest prepareStorage(InputStream fileStream, InputStream cvServiceResponseFileStream, InputStream imageInfoStream) {
@@ -147,6 +161,17 @@ public abstract class AbstractTest {
return buildDefaultLayoutParsingRequest(LayoutParsingType.REDACT_MANAGER);
}
@SneakyThrows
protected LayoutParsingRequest prepareStorage(InputStream fileStream, InputStream cvServiceResponseFileStream, InputStream imageInfoStream, InputStream visualLayoutParsingResponseFileStream) {
storageService.storeObject(TenantContext.getTenantId(), IMAGE_FILE_ID, imageInfoStream);
storageService.storeObject(TenantContext.getTenantId(), TABLE_FILE_ID, cvServiceResponseFileStream);
storageService.storeObject(TenantContext.getTenantId(), ORIGIN_FILE_ID, fileStream);
storageService.storeObject(TenantContext.getTenantId(),VISUAL_LAYOUT_FILE,visualLayoutParsingResponseFileStream );
return buildDefaultLayoutParsingRequest(LayoutParsingType.REDACT_MANAGER);
}
@AfterEach
public void cleanupStorage() {
@@ -10,6 +10,7 @@ import com.knecon.fforesight.service.layoutparser.processor.LayoutParsingPipelin
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationDocument;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.TableServiceResponse;
import com.knecon.fforesight.service.layoutparser.processor.python_api.model.table.VisualLayoutParsingResponse;
import com.knecon.fforesight.service.layoutparser.processor.services.factory.DocumentGraphFactory;
import lombok.SneakyThrows;
@@ -25,7 +26,7 @@ public abstract class BuildDocumentTest extends AbstractTest {
File fileResource = new ClassPathResource(filename).getFile();
prepareStorage(filename);
return layoutParsingPipeline.parseLayout(layoutParsingType, fileResource, layoutParsingStorageService.getImagesFile(IMAGE_FILE_ID), new TableServiceResponse(), filename);
return layoutParsingPipeline.parseLayout(layoutParsingType, fileResource, layoutParsingStorageService.getImagesFile(IMAGE_FILE_ID), new TableServiceResponse(), new VisualLayoutParsingResponse(),filename);
}
@@ -24,20 +24,31 @@ import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Pa
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.AtomicTextBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.TextBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Ruling;
import com.knecon.fforesight.service.layoutparser.processor.utils.DrawingOptions;
import com.knecon.fforesight.service.layoutparser.processor.utils.PdfVisualisationUtility;
import com.knecon.fforesight.service.layoutparser.processor.utils.RectangleTransformations;
import lombok.AccessLevel;
import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Getter;
import lombok.SneakyThrows;
import lombok.experimental.FieldDefaults;
import lombok.experimental.UtilityClass;
@UtilityClass
public class PdfDraw {
public static void drawRectanglesPerPage(String filename, List<List<Rectangle2D>> rectanglesPerPage, String tmpFileName, DrawingOptions options) throws IOException {
ClassPathResource pdfResource = new ClassPathResource(filename);
try (PDDocument pdDocument = Loader.loadPDF(pdfResource.getFile()); var out = new FileOutputStream(tmpFileName)) {
for (int pageNumber = 1; pageNumber < pdDocument.getNumberOfPages() + 1; pageNumber++) {
PdfVisualisationUtility.drawRectangle2DList(pdDocument,
pageNumber,
rectanglesPerPage.get(pageNumber - 1),
options);
}
pdDocument.save(out);
}
}
public static void drawRectanglesPerPage(String filename, List<List<Rectangle2D>> rectanglesPerPage, String tmpFileName) throws IOException {
ClassPathResource pdfResource = new ClassPathResource(filename);
@@ -46,7 +57,7 @@ public class PdfDraw {
PdfVisualisationUtility.drawRectangle2DList(pdDocument,
pageNumber,
rectanglesPerPage.get(pageNumber - 1),
PdfVisualisationUtility.Options.builder().stroke(true).build());
DrawingOptions.builder().stroke(true).build());
}
pdDocument.save(out);
}
@@ -62,13 +73,13 @@ public class PdfDraw {
var rectanglesOnPage = rectanglesPerPage.get(pageNumber - 1);
for (int lineNumber = 0; lineNumber < rectanglesOnPage.size(); lineNumber++) {
var rectanglesInLine = rectanglesOnPage.get(lineNumber);
PdfVisualisationUtility.drawRectangle2DList(pdDocument, pageNumber, rectanglesInLine, PdfVisualisationUtility.Options.builder().stroke(true).build());
PdfVisualisationUtility.drawRectangle2DList(pdDocument, pageNumber, rectanglesInLine, DrawingOptions.builder().stroke(true).build());
double y = Math.min(rectanglesInLine.get(0).getMinY(), rectanglesInLine.get(0).getMaxY());
PdfVisualisationUtility.drawText(String.format("%d", lineNumber),
pdDocument,
new Point2D.Double(rectanglesInLine.get(0).getX() - (5 + (5 * countNumberOfDigits(lineNumber))), y + 2),
pageNumber,
PdfVisualisationUtility.Options.builder().stroke(true).build());
DrawingOptions.builder().stroke(true).build());
}
}
pdDocument.save(out);
@@ -99,20 +110,20 @@ public class PdfDraw {
public static void drawNode(PDDocument document, DocumentTree.Entry entry) {
Options options = buildStandardOptionsForNodes(entry);
DrawingOptions options = buildStandardOptionsForNodes(entry);
drawBBoxAndLabelAndNumberOnPage(document, entry, options);
}
public static void drawTextBlock(PDDocument document, TextBlock textBlock, Options options) {
public static void drawTextBlock(PDDocument document, TextBlock textBlock, DrawingOptions options) {
textBlock.getAtomicTextBlocks().forEach(atb -> drawAtomicTextBlock(document, atb, options));
}
public static void drawAtomicTextBlock(PDDocument document, AtomicTextBlock atomicTextBlock, Options options) {
public static void drawAtomicTextBlock(PDDocument document, AtomicTextBlock atomicTextBlock, DrawingOptions options) {
drawRectangle2DList(document, atomicTextBlock.getPage().getNumber(), atomicTextBlock.getPositions().stream().toList(), options);
@@ -120,7 +131,7 @@ public class PdfDraw {
@SneakyThrows
private static void drawText(String string, PDDocument document, Point2D location, Integer pageNumber, Options options, boolean rotate) {
private static void drawText(String string, PDDocument document, Point2D location, Integer pageNumber, DrawingOptions options, boolean rotate) {
var pdPage = document.getPage(pageNumber - 1);
var contentStream = new PDPageContentStream(document, pdPage, PDPageContentStream.AppendMode.APPEND, true);
@@ -142,14 +153,14 @@ public class PdfDraw {
@SneakyThrows
public static void drawRectangle2DList(PDDocument document, int pageNumber, List<Rectangle2D> rectCollection, Options options) {
public static void drawRectangle2DList(PDDocument document, int pageNumber, List<Rectangle2D> rectCollection, DrawingOptions options) {
var pdPage = document.getPage(pageNumber - 1);
drawRectangle2DList(document, rectCollection, options, pdPage);
}
private static void drawRectangle2DList(PDDocument document, List<Rectangle2D> rectCollection, Options options, PDPage pdPage) throws IOException {
private static void drawRectangle2DList(PDDocument document, List<Rectangle2D> rectCollection, DrawingOptions options, PDPage pdPage) throws IOException {
var contentStream = new PDPageContentStream(document, pdPage, PDPageContentStream.AppendMode.APPEND, true);
@@ -181,12 +192,12 @@ public class PdfDraw {
// PdfVisualisationUtility.drawLine2DList(pdDocument,
// pageNumber,
// list.get(pageNumber - 1),
// PdfVisualisationUtility.Options.builder().stroke(true).build());
// PdfVisualisationUtility.DrawingOptions.builder().stroke(true).build());
PdfVisualisationUtility.drawRectangle2DList(pdDocument,
pageNumber,
rectanglesPerPage.get(pageNumber - 1),
PdfVisualisationUtility.Options.builder().stroke(true).build());
PdfVisualisationUtility.drawRectangle2DList(pdDocument, pageNumber, list.get(pageNumber - 1), PdfVisualisationUtility.Options.builder().stroke(true).build());
DrawingOptions.builder().stroke(true).build());
PdfVisualisationUtility.drawRectangle2DList(pdDocument, pageNumber, list.get(pageNumber - 1), DrawingOptions.builder().stroke(true).build());
}
pdDocument.save(out);
}
@@ -202,35 +213,18 @@ public class PdfDraw {
PdfVisualisationUtility.drawLine2DList(pdDocument,
pageNumber,
linesPerPage.get(pageNumber - 1),
PdfVisualisationUtility.Options.builder().strokeColor(Color.RED).stroke(true).build());
DrawingOptions.builder().strokeColor(Color.RED).stroke(true).build());
}
pdDocument.save(out);
}
}
@Builder
@AllArgsConstructor
@Getter
@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
public static class Options {
boolean stroke;
@Builder.Default
Color strokeColor = Color.BLACK;
@Builder.Default
float strokeWidth = 1f;
boolean fill;
@Builder.Default
Color fillColor = Color.BLACK;
}
private static Options buildStandardOptionsForNodes(DocumentTree.Entry entry) {
private static DrawingOptions buildStandardOptionsForNodes(DocumentTree.Entry entry) {
return Options.builder().stroke(true).strokeColor(switch (entry.getType()) {
return DrawingOptions.builder().stroke(true).strokeColor(switch (entry.getType()) {
case DOCUMENT -> Color.LIGHT_GRAY;
case HEADER, FOOTER -> Color.GREEN;
case PARAGRAPH -> Color.BLUE;
@@ -243,7 +237,7 @@ public class PdfDraw {
}
private static void drawBBoxAndLabelAndNumberOnPage(PDDocument document, DocumentTree.Entry entry, Options options) {
private static void drawBBoxAndLabelAndNumberOnPage(PDDocument document, DocumentTree.Entry entry, DrawingOptions options) {
Map<Page, Rectangle2D> rectanglesPerPage = entry.getNode().getBBox();
for (Page page : rectanglesPerPage.keySet()) {
@@ -0,0 +1,8 @@
{
"dossierId": "123",
"fileId": "123",
"targetFileExtension": "ORIGIN.pdf.gz",
"responseFileExtension": "EXTRACTED_TABLES.json.gz",
"data": [],
"X_TENANT_ID": ""
}
@@ -0,0 +1,19 @@
plugins {
id("com.knecon.fforesight.java-conventions")
id("io.freefair.lombok") version "8.2.2"
}
description = "Library for adding/removing layers in the viewer document"
var pdfBoxVersion = "3.0.0"
dependencies {
implementation("org.apache.pdfbox:pdfbox:${pdfBoxVersion}")
implementation("org.apache.pdfbox:pdfbox-tools:${pdfBoxVersion}")
implementation("org.slf4j:slf4j-api:1.7.25")
implementation("com.knecon.fforesight:tracing-commons:0.5.0")
testImplementation("org.apache.logging.log4j:log4j-slf4j-impl:2.22.1")
testImplementation("org.junit.jupiter:junit-jupiter")
testImplementation platform('org.junit:junit-bom:5.10.0')
}
@@ -0,0 +1,42 @@
package com.knecon.fforesight.service.viewerdoc;
import java.util.List;
import org.apache.pdfbox.cos.COSName;
import lombok.AccessLevel;
import lombok.experimental.FieldDefaults;
@FieldDefaults(makeFinal = true, level = AccessLevel.PUBLIC)
public class ContentStreams {
public static Identifier KNECON_LAYOUT = new Identifier("Layout grid", COSName.getPDFName("KNECON_LAYOUT"), true);
public static Identifier KNECON_VISUAL_PARSING = new Identifier("Layout grid - visual", COSName.getPDFName("KNECON_VISUAL_PARSING"), true);
public static Identifier KNECON_OCR = new Identifier("OCR", COSName.getPDFName("KNECON_OCR"), false);
public static Identifier KNECON_OCR_TEXT_DEBUG = new Identifier("OCR Text", COSName.getPDFName("KNECON_OCR_TEXT_DEBUG"), true);
public static Identifier KNECON_OCR_BBOX_DEBUG = new Identifier("OCR Boxes", COSName.getPDFName("KNECON_OCR_BBOX_DEBUG"), true);
public static Identifier OTHER = new Identifier("other", COSName.getPDFName("OTHER"), false);
public static Identifier ESCAPE_START = new Identifier("escape start", COSName.getPDFName("ESCAPE_START"), false);
public static Identifier ESCAPE_END = new Identifier("escape start", COSName.getPDFName("ESCAPE_END"), false);
public static List<Identifier> allContentStreams = List.of(KNECON_LAYOUT,
KNECON_VISUAL_PARSING,
KNECON_OCR,
KNECON_OCR_BBOX_DEBUG,
KNECON_OCR_TEXT_DEBUG,
OTHER,
ESCAPE_START,
ESCAPE_END);
public record Identifier(String name, COSName cosName, boolean optionalContent) {
}
}
@@ -0,0 +1,8 @@
package com.knecon.fforesight.service.viewerdoc.model;
import java.awt.Color;
import java.awt.geom.Line2D;
public record ColoredLine(Line2D line, Color color, float lineWidth) {
}
@@ -1,8 +1,8 @@
package com.knecon.fforesight.service.layoutparser.processor.model.visualization;
package com.knecon.fforesight.service.viewerdoc.model;
import java.awt.Color;
import java.awt.geom.Rectangle2D;
public record ColoredRectangle(Rectangle2D rectangle2D, Color color) {
public record ColoredRectangle(Rectangle2D rectangle2D, Color color, float lineWidth) {
}
@@ -0,0 +1,10 @@
package com.knecon.fforesight.service.viewerdoc.model;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.apache.pdfbox.pdmodel.font.PDFont;
public interface EmbeddableFont {
PDFont embed(PDDocument document);
}
@@ -1,4 +1,4 @@
package com.knecon.fforesight.service.layoutparser.processor.model.visualization;
package com.knecon.fforesight.service.viewerdoc.model;
import java.awt.Color;
import java.awt.geom.Rectangle2D;
@@ -1,7 +1,7 @@
package com.knecon.fforesight.service.layoutparser.processor.model.visualization;
package com.knecon.fforesight.service.viewerdoc.model;
import java.util.ArrayList;
import java.util.List;
import java.util.HashMap;
import java.util.Map;
import lombok.AccessLevel;
import lombok.Getter;
@@ -12,15 +12,15 @@ import lombok.experimental.FieldDefaults;
public class LayoutGrid {
int numberOfPages;
List<VisualizationsOnPage> visualizationsPerPages;
Map<Integer, VisualizationsOnPage> visualizationsPerPages;
public LayoutGrid(int numberOfPages) {
this.numberOfPages = numberOfPages;
this.visualizationsPerPages = new ArrayList<>(numberOfPages);
this.visualizationsPerPages = new HashMap<>();
for (int i = 0; i < numberOfPages; i++) {
this.visualizationsPerPages.add(VisualizationsOnPage.builder().pageNumber(i).build());
this.visualizationsPerPages.put(i, VisualizationsOnPage.builder().build());
}
}
@@ -0,0 +1,10 @@
package com.knecon.fforesight.service.viewerdoc.model;
import java.util.List;
import org.apache.pdfbox.contentstream.operator.Operator;
import org.apache.pdfbox.cos.COSBase;
public record OperatorWithArguments(Operator operator, List<COSBase> arguments) {
}
@@ -0,0 +1,17 @@
package com.knecon.fforesight.service.viewerdoc.model;
import java.awt.Color;
import java.awt.geom.Point2D;
import java.util.Optional;
import org.apache.pdfbox.pdmodel.graphics.state.RenderingMode;
import org.apache.pdfbox.util.Matrix;
public record PlacedText(String text, Point2D lineStart, Color color, float fontSize, EmbeddableFont font, Optional<Matrix> textMatrix, Optional<RenderingMode> renderingMode) {
public static PlacedText textFacingUp(String text, Point2D lineStart, float fontSize, Color color, EmbeddableFont font) {
return new PlacedText(text, lineStart, color, fontSize, font, Optional.empty(), Optional.empty());
}
}
@@ -0,0 +1,37 @@
package com.knecon.fforesight.service.viewerdoc.model;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.apache.pdfbox.pdmodel.font.PDFont;
import org.apache.pdfbox.pdmodel.font.PDType1Font;
import org.apache.pdfbox.pdmodel.font.Standard14Fonts;
import lombok.RequiredArgsConstructor;
import lombok.SneakyThrows;
@RequiredArgsConstructor
public class Standard14EmbeddableFont implements EmbeddableFont {
private final PDType1Font font;
public static Standard14EmbeddableFont helvetica() {
return new Standard14EmbeddableFont(new PDType1Font(Standard14Fonts.FontName.HELVETICA));
}
@SneakyThrows
public float getStringWidth(String text) {
return font.getStringWidth(text);
}
@Override
public PDFont embed(PDDocument document) {
// no need to embed anything
return font;
}
}
@@ -0,0 +1,23 @@
package com.knecon.fforesight.service.viewerdoc.model;
import java.util.Map;
import com.knecon.fforesight.service.viewerdoc.ContentStreams;
import lombok.AccessLevel;
import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Getter;
import lombok.experimental.FieldDefaults;
@Builder
@Getter
@AllArgsConstructor
@FieldDefaults(level = AccessLevel.PRIVATE)
public class Visualizations {
ContentStreams.Identifier layer;
Map<Integer, VisualizationsOnPage> visualizationsOnPages;
boolean layerVisibilityDefaultValue;
}
@@ -1,4 +1,4 @@
package com.knecon.fforesight.service.layoutparser.processor.model.visualization;
package com.knecon.fforesight.service.viewerdoc.model;
import java.util.LinkedList;
import java.util.List;
@@ -13,7 +13,7 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
public class VisualizationsOnPage {
int pageNumber;
boolean makePathsInvisible;
@Builder.Default
List<PlacedText> placedTexts = new LinkedList<>();
@Builder.Default
@@ -0,0 +1,7 @@
package com.knecon.fforesight.service.viewerdoc.pdf;
import com.knecon.fforesight.service.viewerdoc.ContentStreams;
public record ClassifiedContentStream(SinglePDContentStream contentStream, ContentStreams.Identifier classification) {
}
@@ -0,0 +1,61 @@
package com.knecon.fforesight.service.viewerdoc.pdf;
import java.io.IOException;
import java.io.InputStream;
import org.apache.pdfbox.contentstream.PDContentStream;
import org.apache.pdfbox.io.RandomAccessInputStream;
import org.apache.pdfbox.io.RandomAccessRead;
import org.apache.pdfbox.pdmodel.PDResources;
import org.apache.pdfbox.pdmodel.common.PDRectangle;
import org.apache.pdfbox.pdmodel.common.PDStream;
import org.apache.pdfbox.util.Matrix;
import lombok.AccessLevel;
import lombok.Getter;
import lombok.RequiredArgsConstructor;
import lombok.experimental.FieldDefaults;
@Getter
@RequiredArgsConstructor
@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
public class SinglePDContentStream implements PDContentStream {
PDStream pdStream;
@Override
public InputStream getContents() throws IOException {
return new RandomAccessInputStream(getContentsForRandomAccess());
}
@Override
public RandomAccessRead getContentsForRandomAccess() throws IOException {
return pdStream.getCOSObject().createView();
}
@Override
public PDResources getResources() {
return null;
}
@Override
public PDRectangle getBBox() {
return null;
}
@Override
public Matrix getMatrix() {
return null;
}
}
@@ -0,0 +1,121 @@
package com.knecon.fforesight.service.viewerdoc.service;
import java.util.LinkedList;
import java.util.List;
import java.util.Optional;
import org.apache.pdfbox.contentstream.PDContentStream;
import org.apache.pdfbox.contentstream.operator.OperatorName;
import org.apache.pdfbox.cos.COSName;
import org.apache.pdfbox.pdmodel.PDPage;
import com.knecon.fforesight.service.viewerdoc.ContentStreams;
import com.knecon.fforesight.service.viewerdoc.model.OperatorWithArguments;
import com.knecon.fforesight.service.viewerdoc.pdf.ClassifiedContentStream;
import com.knecon.fforesight.service.viewerdoc.pdf.SinglePDContentStream;
import lombok.SneakyThrows;
import lombok.experimental.UtilityClass;
@UtilityClass
public class ContentStreamClassifier {
public List<ClassifiedContentStream> getClassifiedContentStreams(PDPage page) {
List<SinglePDContentStream> streams = new LinkedList<>();
page.getContentStreams().forEachRemaining(stream -> streams.add(new SinglePDContentStream(stream)));
return ContentStreamClassifier.classifySingleContentStreams(page, streams);
}
public List<ClassifiedContentStream> classifySingleContentStreams(PDPage page, List<SinglePDContentStream> streams) {
return streams.stream().map(singlePDContentStream -> classifySingleContentStream(page, singlePDContentStream)).toList();
}
private ClassifiedContentStream classifySingleContentStream(PDPage page, SinglePDContentStream singlePDContentStream) {
ContentStreams.Identifier classification = classifyContentStream(singlePDContentStream, page);
return new ClassifiedContentStream(singlePDContentStream, classification);
}
/**
* We assume all of our layers are written escaped, so only unknown content streams need to be escaped.
*
* @param classifiers List of all content streams of a page with their classification
* @return false, if any content stream with classification other is not prefixed with an ESCAPE_START and suffixed with an ESCAPE_END
*/
public boolean areAllContentStreamsEscaped(List<ClassifiedContentStream> classifiers) {
int escapeDepth = 0;
for (ClassifiedContentStream classifier : classifiers) {
if (classifier.classification().equals(ContentStreams.OTHER) && escapeDepth == 0) {
return false;
}
if (classifier.classification().equals(ContentStreams.ESCAPE_START)) {
escapeDepth++;
}
if (classifier.classification().equals(ContentStreams.ESCAPE_END)) {
escapeDepth--;
}
}
return escapeDepth == 0;
}
@SneakyThrows
public ContentStreams.Identifier classifyContentStream(PDContentStream contentStream, PDPage page) {
List<OperatorWithArguments> operatorsWithArguments = ContentStreamUtility.parseLeadingOperators(contentStream, 2);
if (operatorsWithArguments.isEmpty()) {
return ContentStreams.OTHER;
}
OperatorWithArguments firstOperator = operatorsWithArguments.get(0);
// If we wrap the content streams we append and prepend a content stream with exactly one operator "q" or "Q".
if (operatorsWithArguments.size() == 1) {
if (firstOperator.operator().getName().equals(OperatorName.SAVE)) {
return ContentStreams.ESCAPE_START;
}
if (firstOperator.operator().getName().equals(OperatorName.RESTORE)) {
return ContentStreams.ESCAPE_END;
}
}
// In previous versions we did not set a marked content with an explicit name. Instead, we wrote an optional content group (OCG) with the name "Layout grid".
// This OCG is then assigned a COSName by PDFBox. Usually its "oc1".
// Thus, in order to find this name we need to look in the page resources to find the COSName assigned to the OCG.
// This COSName can then be found as an argument for the first operator in the content stream.
if (firstOperator.operator().getName().equals(OperatorName.BEGIN_MARKED_CONTENT_SEQ)) {
Optional<COSName> layoutGridOCGName = ContentStreamUtility.findLayoutGridOCGName(page);
if (layoutGridOCGName.isPresent()) {
if (arumentsContainLayoutGridOCG(firstOperator, layoutGridOCGName.get())) {
return ContentStreams.KNECON_LAYOUT;
}
}
}
if (!firstOperator.operator().getName().equals(OperatorName.BEGIN_MARKED_CONTENT)) {
return ContentStreams.OTHER;
}
Optional<COSName> firstCOSNameFromArguments = firstOperator.arguments().stream().filter(c -> c instanceof COSName).map(c -> (COSName) c).findFirst();
if (firstCOSNameFromArguments.isEmpty()) {
return ContentStreams.OTHER;
}
var cosName = firstCOSNameFromArguments.get();
return ContentStreams.allContentStreams.stream().filter(identifier -> identifier.cosName().equals(cosName)).findAny().orElse(ContentStreams.OTHER);
}
private static boolean arumentsContainLayoutGridOCG(OperatorWithArguments operator, COSName layoutGridOCGName) {
return operator.arguments().stream().filter(c -> c instanceof COSName).map(c -> (COSName) c).anyMatch(cosName -> cosName.equals(layoutGridOCGName));
}
}
@@ -0,0 +1,78 @@
package com.knecon.fforesight.service.viewerdoc.service;
import java.io.IOException;
import java.util.ArrayList;
import java.util.LinkedList;
import java.util.List;
import java.util.Optional;
import java.util.Set;
import org.apache.pdfbox.contentstream.PDContentStream;
import org.apache.pdfbox.contentstream.operator.Operator;
import org.apache.pdfbox.cos.COSBase;
import org.apache.pdfbox.cos.COSName;
import org.apache.pdfbox.cos.COSString;
import org.apache.pdfbox.pdfparser.PDFStreamParser;
import org.apache.pdfbox.pdmodel.PDPage;
import org.apache.pdfbox.pdmodel.common.PDStream;
import com.knecon.fforesight.service.viewerdoc.ContentStreams;
import com.knecon.fforesight.service.viewerdoc.model.OperatorWithArguments;
import com.knecon.fforesight.service.viewerdoc.pdf.ClassifiedContentStream;
import com.knecon.fforesight.service.viewerdoc.pdf.SinglePDContentStream;
import lombok.experimental.UtilityClass;
@UtilityClass
public class ContentStreamUtility {
public static List<OperatorWithArguments> parseLeadingOperators(PDContentStream contentStream,
int numberOfOperatorsToRead) throws IOException {
List<COSBase> arguments = new ArrayList<>();
PDFStreamParser parser = new PDFStreamParser(contentStream);
List<OperatorWithArguments> operatorsWithArguments = new LinkedList<>();
for (int i = 0; i < numberOfOperatorsToRead; ) {
Object token = parser.parseNextToken();
if (token == null) {
break;
}
if (token instanceof Operator operator) {
operatorsWithArguments.add(new OperatorWithArguments(operator, arguments));
arguments = new ArrayList<>();
i++;
} else {
arguments.add((COSBase) token);
}
}
return operatorsWithArguments;
}
public static Optional<COSName> findLayoutGridOCGName(PDPage page) {
Optional<COSName> layoutGridOCGName = Optional.empty();
var resourceIterator = page.getResources().getPropertiesNames();
for (COSName cosName : resourceIterator) {
COSBase cosBase = page.getResources().getProperties(cosName).getCOSObject().getDictionaryObject(COSName.NAME);
if (cosBase instanceof COSString string) {
if (ContentStreams.KNECON_LAYOUT.name().equals(string.getString())) {
layoutGridOCGName = Optional.of(cosName);
}
}
}
return layoutGridOCGName;
}
public static List<PDStream> removeLayerFromContentStreams(Set<ContentStreams.Identifier> layers, List<ClassifiedContentStream> classifiers) {
return classifiers.stream()
.filter(classifiedContentStream -> !layers.contains(classifiedContentStream.classification()))
.map(ClassifiedContentStream::contentStream)
.map(SinglePDContentStream::getPdStream)
.toList();
}
}
@@ -0,0 +1,316 @@
package com.knecon.fforesight.service.viewerdoc.service;
import java.awt.geom.AffineTransform;
import java.awt.geom.Rectangle2D;
import java.io.File;
import java.io.FileOutputStream;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.StandardCopyOption;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Optional;
import java.util.Set;
import java.util.stream.Collectors;
import org.apache.pdfbox.Loader;
import org.apache.pdfbox.cos.COSName;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.apache.pdfbox.pdmodel.PDDocumentCatalog;
import org.apache.pdfbox.pdmodel.PDPage;
import org.apache.pdfbox.pdmodel.PDPageContentStream;
import org.apache.pdfbox.pdmodel.PDResources;
import org.apache.pdfbox.pdmodel.font.PDFont;
import org.apache.pdfbox.pdmodel.graphics.optionalcontent.PDOptionalContentGroup;
import org.apache.pdfbox.pdmodel.graphics.optionalcontent.PDOptionalContentProperties;
import org.apache.pdfbox.pdmodel.graphics.state.PDExtendedGraphicsState;
import org.apache.pdfbox.pdmodel.graphics.state.RenderingMode;
import org.apache.pdfbox.util.Matrix;
import com.knecon.fforesight.service.viewerdoc.ContentStreams;
import com.knecon.fforesight.service.viewerdoc.model.ColoredLine;
import com.knecon.fforesight.service.viewerdoc.model.ColoredRectangle;
import com.knecon.fforesight.service.viewerdoc.model.FilledRectangle;
import com.knecon.fforesight.service.viewerdoc.model.PlacedText;
import com.knecon.fforesight.service.viewerdoc.model.Visualizations;
import com.knecon.fforesight.service.viewerdoc.model.VisualizationsOnPage;
import com.knecon.fforesight.service.viewerdoc.pdf.ClassifiedContentStream;
import io.micrometer.observation.Observation;
import io.micrometer.observation.ObservationRegistry;
import io.micrometer.observation.annotation.Observed;
import lombok.RequiredArgsConstructor;
import lombok.SneakyThrows;
import lombok.extern.slf4j.Slf4j;
@Slf4j
@RequiredArgsConstructor
public class ViewerDocumentService {
private final ObservationRegistry registry;
public void addVisualizationsOnPage(File originFile, File destinationFile, Visualizations visualizations) {
addVisualizationsOnPage(originFile, destinationFile, List.of(visualizations));
}
@Observed(name = "ViewerDocumentService", contextualName = "add-visualizations")
@SneakyThrows
public void addVisualizationsOnPage(File originFile, File destinationFile, List<Visualizations> visualizations) {
// originFile and destinationFile might be the same, so we use a temp file.
// Otherwise, saving the document might corrupt the file
Path tmpFile = Files.createTempFile("tmpViewerDocument", ".pdf");
Files.copy(originFile.toPath(), tmpFile, StandardCopyOption.REPLACE_EXISTING);
PDDocument pdDocument = openPDDocument(tmpFile.toFile());
enrichObservation(pdDocument, visualizations.stream().map(Visualizations::getLayer).toList());
Set<ContentStreams.Identifier> allLayers = visualizations.stream().map(Visualizations::getLayer).collect(Collectors.toUnmodifiableSet());
Map<ContentStreams.Identifier, PDOptionalContentGroup> optionalContentGroupMap = addLayersToDocument(visualizations, pdDocument);
for (int pageNumber = 0; pageNumber < pdDocument.getNumberOfPages(); pageNumber++) {
PDPage pdPage = pdDocument.getPage(pageNumber);
createPageResourcesIfNotPresent(pdPage); // needed for optionalContentGroups
List<ClassifiedContentStream> classifiers = ContentStreamClassifier.getClassifiedContentStreams(pdPage);
pdPage.setContents(ContentStreamUtility.removeLayerFromContentStreams(allLayers, classifiers));
AffineTransform textDeRotationMatrix = getTextDeRotationTransform(pdPage);
if (!ContentStreamClassifier.areAllContentStreamsEscaped(classifiers)) {
// We need to save the graphics state before, such that our appended content cannot be affected by previous content streams with side effects,
// e.g. not escaped matrix transformations.
wrapContentStreams(pdDocument, pdPage);
}
for (Visualizations visualization : visualizations) {
if (!visualization.getVisualizationsOnPages().containsKey(pageNumber)) {
continue;
}
// We need to append to the content stream, otherwise the content could be overlapped by following content.
try (var contentStream = new PDPageContentStream(pdDocument, pdPage, PDPageContentStream.AppendMode.APPEND, true)) {
contentStream.beginMarkedContent(visualization.getLayer().cosName());
if (optionalContentGroupMap.containsKey(visualization.getLayer())) {
contentStream.beginMarkedContent(COSName.OC, optionalContentGroupMap.get(visualization.getLayer()));
}
contentStream.saveGraphicsState();
drawVisualizationsToContentStream(pdDocument, visualization.getVisualizationsOnPages().get(pageNumber), contentStream, textDeRotationMatrix);
contentStream.restoreGraphicsState();
if (optionalContentGroupMap.containsKey(visualization.getLayer())) {
contentStream.endMarkedContent();
}
contentStream.endMarkedContent();
}
}
if (pageNumber % 500 == 0 && pageNumber != 0) { // re-open document every once in a while to save on RAM
log.info("Incremental save after {}/{} pages", pageNumber, pdDocument.getNumberOfPages());
observedIncrementalSave(pdDocument, destinationFile);
pdDocument.close();
Files.copy(destinationFile.toPath(), tmpFile, StandardCopyOption.REPLACE_EXISTING);
pdDocument = openPDDocument(tmpFile.toFile());
}
}
observedIncrementalSave(pdDocument, destinationFile);
pdDocument.close();
assert tmpFile.toFile().delete();
}
private static Map<ContentStreams.Identifier, PDOptionalContentGroup> addLayersToDocument(List<Visualizations> visualizations, PDDocument pdDocument) {
Map<ContentStreams.Identifier, PDOptionalContentGroup> optionalContentGroupMap = new HashMap<>();
for (Visualizations visualization : visualizations) {
addLayerToDocument(visualization.getLayer(), pdDocument, visualization.isLayerVisibilityDefaultValue())//
.ifPresent(ocg -> optionalContentGroupMap.put(visualization.getLayer(), ocg));
}
return optionalContentGroupMap;
}
private static void drawVisualizationsToContentStream(PDDocument pdDocument,
VisualizationsOnPage visualizationsOnPage,
PDPageContentStream contentStream,
AffineTransform textDeRotationMatrix) throws IOException {
if (visualizationsOnPage.isMakePathsInvisible()) {
contentStream.addRect(0, 0, 1, 1);
contentStream.clip();
}
for (ColoredLine coloredLine : visualizationsOnPage.getColoredLines()) {
contentStream.setLineWidth(coloredLine.lineWidth());
contentStream.setStrokingColor(coloredLine.color());
contentStream.moveTo((float) coloredLine.line().getX1(), (float) coloredLine.line().getY1());
contentStream.lineTo((float) coloredLine.line().getX2(), (float) coloredLine.line().getY2());
contentStream.stroke();
}
for (ColoredRectangle coloredRectangle : visualizationsOnPage.getColoredRectangles()) {
contentStream.setLineWidth(coloredRectangle.lineWidth());
contentStream.setStrokingColor(coloredRectangle.color());
Rectangle2D r = coloredRectangle.rectangle2D();
contentStream.addRect((float) r.getX(), (float) r.getY(), (float) r.getWidth(), (float) r.getHeight());
contentStream.stroke();
}
for (FilledRectangle filledRectangle : visualizationsOnPage.getFilledRectangles()) {
contentStream.setNonStrokingColor(filledRectangle.color());
PDExtendedGraphicsState graphicsState = new PDExtendedGraphicsState();
graphicsState.setNonStrokingAlphaConstant(filledRectangle.alpha());
contentStream.setGraphicsStateParameters(graphicsState);
Rectangle2D r = filledRectangle.rectangle2D();
contentStream.addRect((float) r.getX(), (float) r.getY(), (float) r.getWidth(), (float) r.getHeight());
contentStream.fill();
}
for (PlacedText placedText : visualizationsOnPage.getPlacedTexts()) {
PDFont font = placedText.font().embed(pdDocument);
contentStream.setFont(font, placedText.fontSize());
contentStream.beginText();
contentStream.setNonStrokingColor(placedText.color());
if (placedText.renderingMode().isPresent()) {
contentStream.setRenderingMode(placedText.renderingMode().get());
} else {
contentStream.setRenderingMode(RenderingMode.FILL);
}
Matrix textMatrix = getTextMatrix(placedText, textDeRotationMatrix);
contentStream.setTextMatrix(textMatrix);
contentStream.showText(placedText.text());
contentStream.endText();
}
}
private void enrichObservation(PDDocument pdDocument, List<ContentStreams.Identifier> layers) {
if (registry == null || registry.getCurrentObservation() == null || registry.isNoop()) {
return;
}
registry.getCurrentObservation().highCardinalityKeyValue("numberOfPages", String.valueOf(pdDocument.getNumberOfPages()));
for (int i = 0; i < layers.size(); i++) {
ContentStreams.Identifier layer = layers.get(i);
registry.getCurrentObservation().highCardinalityKeyValue("layer_" + i, String.valueOf(layer.name()));
}
}
private static void wrapContentStreams(PDDocument pdDocument, PDPage pdPage) throws IOException {
try (var contentStream = new PDPageContentStream(pdDocument, pdPage, PDPageContentStream.AppendMode.PREPEND, false)) {
contentStream.saveGraphicsState();
}
try (var contentStream = new PDPageContentStream(pdDocument, pdPage, PDPageContentStream.AppendMode.APPEND, false)) {
contentStream.restoreGraphicsState();
}
}
private static Matrix getTextMatrix(PlacedText placedText, AffineTransform textDeRotationMatrix) {
Matrix textMatrix;
if (placedText.textMatrix().isEmpty()) {
textMatrix = new Matrix((float) textDeRotationMatrix.getScaleX(),
(float) textDeRotationMatrix.getShearX(),
(float) textDeRotationMatrix.getShearY(),
(float) textDeRotationMatrix.getScaleY(),
(float) placedText.lineStart().getX(),
(float) placedText.lineStart().getY());
} else {
textMatrix = placedText.textMatrix().get();
}
return textMatrix;
}
private static Optional<PDOptionalContentGroup> addLayerToDocument(ContentStreams.Identifier layer, PDDocument pdDocument, boolean layerVisibilityDefaultValue) {
if (layer.optionalContent()) {
return Optional.of(addLayerToDocument(pdDocument, layer.name(), layerVisibilityDefaultValue));
}
return Optional.empty();
}
private static PDOptionalContentGroup addLayerToDocument(PDDocument pdDocument, String layerName, boolean layerVisibilityDefaultValue) {
PDDocumentCatalog catalog = pdDocument.getDocumentCatalog();
PDOptionalContentProperties ocprops = catalog.getOCProperties();
if (ocprops == null) {
ocprops = new PDOptionalContentProperties();
catalog.setOCProperties(ocprops);
}
PDOptionalContentGroup layer = null;
if (ocprops.hasGroup(layerName)) {
layer = ocprops.getGroup(layerName);
} else {
layer = new PDOptionalContentGroup(layerName);
ocprops.addGroup(layer);
}
ocprops.setGroupEnabled(layer, layerVisibilityDefaultValue);
return layer;
}
private static PDDocument openPDDocument(File tmpFile) throws IOException {
PDDocument pdDocument;
pdDocument = Loader.loadPDF(tmpFile);
pdDocument.setAllSecurityToBeRemoved(true);
return pdDocument;
}
@SneakyThrows
private void observedIncrementalSave(PDDocument pdDocument, File outputFile) {
Observation.createNotStarted("ViewerDocumentService", registry).contextualName("incremental-save").observe(() -> {
try (var out = new FileOutputStream(outputFile)) {
pdDocument.save(out);
} catch (IOException e) {
throw new RuntimeException(e);
}
});
}
private static void createPageResourcesIfNotPresent(PDPage pdPage) {
PDResources resources = pdPage.getResources();
if (resources == null) {
resources = new PDResources();
pdPage.setResources(resources);
}
}
private static AffineTransform getTextDeRotationTransform(PDPage page) {
return AffineTransform.getQuadrantRotateInstance(switch (page.getRotation()) {
case 90 -> 3;
case 180 -> 2;
case 270 -> 1;
default -> 0;
});
}
}
@@ -0,0 +1,124 @@
package com.knecon.fforesight.service.viewerdoc.service;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertTrue;
import java.io.File;
import java.nio.file.Files;
import java.util.List;
import java.util.Set;
import java.util.stream.Collectors;
import org.apache.pdfbox.Loader;
import org.apache.pdfbox.cos.COSName;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.apache.pdfbox.pdmodel.PDPage;
import org.junit.jupiter.api.Test;
import com.knecon.fforesight.service.viewerdoc.ContentStreams;
import com.knecon.fforesight.service.viewerdoc.pdf.ClassifiedContentStream;
import lombok.SneakyThrows;
import lombok.extern.slf4j.Slf4j;
@Slf4j
public class ContentStreamClassifierTest {
@Test
@SneakyThrows
public void testClassification() {
File pdfFile = new File(Thread.currentThread().getContextClassLoader().getResource("viewerDocLayers.pdf").getFile());
try (PDDocument document = Loader.loadPDF(pdfFile)) {
PDPage page = document.getPage(0);
List<ClassifiedContentStream> classifieds = ContentStreamClassifier.getClassifiedContentStreams(page);
logContentStreamClassifications(classifieds);
assertEquals(11, classifieds.size());
assertEquals(ContentStreams.ESCAPE_START, classifieds.get(0).classification());
for (int i = 1; i < 9; i++) {
assertEquals(ContentStreams.OTHER, classifieds.get(i).classification());
}
assertEquals(ContentStreams.ESCAPE_END, classifieds.get(9).classification());
assertEquals(ContentStreams.KNECON_LAYOUT, classifieds.get(10).classification());
assertTrue(ContentStreamClassifier.areAllContentStreamsEscaped(classifieds));
}
}
@Test
@SneakyThrows
public void testRemoveLayoutLayer() {
File pdfFile = new File(Thread.currentThread().getContextClassLoader().getResource("viewerDocLayers.pdf").getFile());
File tmpFile = Files.createTempFile("removedLayout", ".pdf").toFile();
try (PDDocument document = Loader.loadPDF(pdfFile)) {
PDPage page = document.getPage(0);
List<ClassifiedContentStream> classifieds = ContentStreamClassifier.getClassifiedContentStreams(page);
page.setContents(ContentStreamUtility.removeLayerFromContentStreams(Set.of(ContentStreams.KNECON_LAYOUT), classifieds));
document.save(tmpFile);
}
try (PDDocument document2 = Loader.loadPDF(tmpFile)) {
PDPage page2 = document2.getPage(0);
List<ClassifiedContentStream> classifieds2 = ContentStreamClassifier.getClassifiedContentStreams(page2);
logContentStreamClassifications(classifieds2);
assertEquals(10, classifieds2.size());
assertEquals(ContentStreams.ESCAPE_START, classifieds2.get(0).classification());
for (int i = 1; i < 9; i++) {
assertEquals(ContentStreams.OTHER, classifieds2.get(i).classification());
}
assertEquals(ContentStreams.ESCAPE_END, classifieds2.get(9).classification());
assertTrue(ContentStreamClassifier.areAllContentStreamsEscaped(classifieds2));
}
assert tmpFile.delete();
}
@Test
@SneakyThrows
public void testClassificationForOldLayers() {
File pdfFile = new File(Thread.currentThread().getContextClassLoader().getResource("oldViewerDocLayers.pdf").getFile());
try (PDDocument document = Loader.loadPDF(pdfFile)) {
PDPage page = document.getPage(0);
List<ClassifiedContentStream> classifieds = ContentStreamClassifier.getClassifiedContentStreams(page);
logContentStreamClassifications(classifieds);
assertEquals(11, classifieds.size());
assertEquals(ContentStreams.ESCAPE_START, classifieds.get(0).classification());
for (int i = 1; i < 9; i++) {
assertEquals(ContentStreams.OTHER, classifieds.get(i).classification());
}
assertEquals(ContentStreams.ESCAPE_END, classifieds.get(9).classification());
assertEquals(ContentStreams.KNECON_LAYOUT, classifieds.get(10).classification());
assertTrue(ContentStreamClassifier.areAllContentStreamsEscaped(classifieds));
}
}
private static void logContentStreamClassifications(List<ClassifiedContentStream> classifieds) {
log.info("number of content streams: {}", classifieds.size());
log.info("Classifications: {}", classifieds.stream()//
.map(ClassifiedContentStream::classification)//
.map(ContentStreams.Identifier::cosName)//
.map(COSName::getName)//
.collect(Collectors.joining(", ")));
}
}
@@ -0,0 +1,16 @@
<Configuration>
<Appenders>
<Console name="CONSOLE" target="SYSTEM_OUT">
<PatternLayout pattern="%d{HH:mm:ss.SSS} [%t] %-5level %logger{36} - %msg%n"/>
</Console>
</Appenders>
<Loggers>
<Root level="warn">
<AppenderRef ref="CONSOLE"/>
</Root>
<Logger name="com.knecon" level="info"/>
</Loggers>
</Configuration>
+2 -6
View File
@@ -1,13 +1,9 @@
/*
* This file was generated by the Gradle 'init' task.
*
* This project uses @Incubating APIs which are subject to change.
*/
rootProject.name = "layoutparser"
include(":layoutparser-service-server")
include(":layoutparser-service-processor")
include(":layoutparser-service-internal-api")
include("viewer-doc-processor")
project(":layoutparser-service-server").projectDir = file("layoutparser-service/layoutparser-service-server")
project(":layoutparser-service-processor").projectDir = file("layoutparser-service/layoutparser-service-processor")
project(":layoutparser-service-internal-api").projectDir = file("layoutparser-service/layoutparser-service-internal-api")
project(":viewer-doc-processor").projectDir = file("layoutparser-service/viewer-doc-processor")