Compare commits

..
10 Commits
Author SHA1 Message Date
Dominique Eifländer 08e994d904 More refactoring 2024-02-20 14:04:28 +01:00
Dominique Eifländer 0c8c727303 Reading Order 2024-02-20 11:56:43 +01:00
Dominique Eifländer 72202f63dc More 2024-02-16 14:51:26 +01:00
Dominique Eifländer e14d953b04 More 2024-02-16 14:50:31 +01:00
Dominique Eifländer 9e5778d4b2 More 2024-02-16 14:08:59 +01:00
Dominique Eifländer e394f2fa7c More refactoring 2024-02-16 13:48:03 +01:00
Dominique Eifländer b2fb6829cb More refactoring 2024-02-16 11:15:44 +01:00
Dominique Eifländer 4871e55f2d More refactoring 2024-02-15 16:54:07 +01:00
Dominique Eifländer 4de6c12aec REmove more 2024-02-15 10:29:39 +01:00
Dominique Eifländer 4afa8daafa First working docstrum 2024-02-14 13:39:27 +01:00
62 changed files with 1844 additions and 824 deletions
@@ -18,15 +18,11 @@ 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.")//
@@ -10,6 +10,7 @@ import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.atomic.AtomicReference;
import org.apache.pdfbox.Loader;
@@ -26,6 +27,7 @@ import com.knecon.fforesight.service.layoutparser.internal.api.queue.LayoutParsi
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.ClassificationPage;
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationSection;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.model.image.ClassifiedImage;
import com.knecon.fforesight.service.layoutparser.processor.model.table.CleanRulings;
@@ -33,11 +35,9 @@ 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;
@@ -49,6 +49,7 @@ import com.knecon.fforesight.service.layoutparser.processor.services.blockificat
import com.knecon.fforesight.service.layoutparser.processor.services.classification.DocuMineClassificationService;
import com.knecon.fforesight.service.layoutparser.processor.services.classification.RedactManagerClassificationService;
import com.knecon.fforesight.service.layoutparser.processor.services.classification.TaasClassificationService;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.DocstrumSegmentationService;
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.TaasDocumentDataMapper;
@@ -88,7 +89,7 @@ public class LayoutParsingPipeline {
RedactManagerBlockificationService redactManagerBlockificationService;
LayoutGridService layoutGridService;
ObservationRegistry observationRegistry;
VisualLayoutParsingAdapter visualLayoutParsingAdapter;
DocstrumSegmentationService docstrumSegmentationService;
public LayoutParsingFinishedEvent parseLayoutAndSaveFilesToStorage(LayoutParsingRequest layoutParsingRequest) throws IOException {
@@ -97,33 +98,22 @@ public class LayoutParsingPipeline {
log.info("Starting layout parsing for {}", layoutParsingRequest.identifier());
File originFile = layoutParsingStorageService.getOriginFile(layoutParsingRequest.originFileStorageId());
File viewerDocumentFile = layoutParsingStorageService.getViewerDocFile(layoutParsingRequest.viewerDocumentStorageId())
.orElse(originFile);
VisualLayoutParsingResponse visualLayoutParsingResponse = new VisualLayoutParsingResponse();
if (layoutParsingRequest.visualLayoutParsingFileId().isPresent()) {
visualLayoutParsingResponse = layoutParsingStorageService.getVisualLayoutParsingFile(layoutParsingRequest.visualLayoutParsingFileId().get());
}
File viewerDocumentFile = layoutParsingStorageService.getViewerDocFile(layoutParsingRequest.viewerDocumentStorageId()).orElse(originFile);
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,
visualLayoutParsingResponse,
originFile,
imageServiceResponse,
tableServiceResponse,
layoutParsingRequest.identifier().toString());
log.info("Building document graph for {}", layoutParsingRequest.identifier());
@@ -156,25 +146,25 @@ public class LayoutParsingPipeline {
.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();
}
@@ -184,9 +174,9 @@ 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();
}
@@ -195,14 +185,14 @@ 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));
}
@@ -212,17 +202,12 @@ 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 = new HashMap<>();
if(signatures.size() > 0) {
visualLayoutParsingAdapter.buildExtractedSignaturesPerPage(visualLayoutParsingResponse);
}
ClassificationDocument classificationDocument = new ClassificationDocument();
List<ClassificationPage> classificationPages = new ArrayList<>();
@@ -262,11 +247,37 @@ public class LayoutParsingPipeline {
PDRectangle cropbox = pdPage.getCropBox();
CleanRulings cleanRulings = rulingCleaningService.getCleanRulings(pdfTableCells.get(pageNumber), stripper.getRulings());
ClassificationPage classificationPage = switch (layoutParsingType) {
case REDACT_MANAGER -> redactManagerBlockificationService.blockify(stripper.getTextPositionSequences(), cleanRulings.getHorizontal(), cleanRulings.getVertical());
case TAAS -> taasBlockificationService.blockify(stripper.getTextPositionSequences(), cleanRulings.getHorizontal(), cleanRulings.getVertical());
case DOCUMINE -> docuMineBlockificationService.blockify(stripper.getTextPositionSequences(), cleanRulings.getHorizontal(), cleanRulings.getVertical());
};
// Docstrum
AtomicInteger num = new AtomicInteger(pageNumber);
var zones = docstrumSegmentationService.segmentPage(stripper.getTextPositionSequences());
List<AbstractPageBlock> pageBlocks = new ArrayList<>();
AtomicInteger numOnPage = new AtomicInteger(1);
// List<TextPositionSequence> textPositionSequences = new ArrayList<>();
zones.forEach(zone -> {
List<TextPositionSequence> textPositionSequences = new ArrayList<>();
zone.getLines().forEach(line -> {
line.getWords().forEach(word -> {
textPositionSequences.add(new TextPositionSequence(word.getTextPositions(), num.get()));
});
});
var cps = redactManagerBlockificationService.blockify(textPositionSequences, cleanRulings.getHorizontal(), cleanRulings.getVertical());
cps.getTextBlocks().forEach(cp -> {
pageBlocks.add(redactManagerBlockificationService.buildTextBlock(((TextPageBlock) cp).getSequences(), numOnPage.getAndIncrement()));
});
});
// ClassificationPage classificationPage = switch (layoutParsingType) {
// case REDACT_MANAGER -> redactManagerBlockificationService.blockify(textPositionSequences, cleanRulings.getHorizontal(), cleanRulings.getVertical());
// case TAAS -> taasBlockificationService.blockify(textPositionSequences, cleanRulings.getHorizontal(), cleanRulings.getVertical());
// case DOCUMINE -> docuMineBlockificationService.blockify(textPositionSequences, cleanRulings.getHorizontal(), cleanRulings.getVertical());
// };
ClassificationPage classificationPage = new ClassificationPage(pageBlocks);
classificationPage.setCleanRulings(cleanRulings);
classificationPage.setRotation(rotation);
classificationPage.setLandscape(isLandscape);
@@ -283,10 +294,6 @@ public class LayoutParsingPipeline {
imageServiceResponseAdapter.findOcr(classificationPage);
}
if(signatures.containsKey(pageNumber)) {
classificationPage.setImages(signatures.get(pageNumber));
}
tableExtractionService.extractTables(cleanRulings, classificationPage);
buildPageStatistics(classificationPage);
@@ -306,9 +313,19 @@ public class LayoutParsingPipeline {
case REDACT_MANAGER -> redactManagerClassificationService.classifyDocument(classificationDocument);
}
List<ClassificationSection> sections = new ArrayList<>();
for (var page : classificationPages) {
page.getTextBlocks().forEach(block -> {
block.setPage(page.getPageNumber());
var section = sectionsBuilderService.buildTextBlock(List.of(block), "a");
sections.add(section);
});
}
classificationDocument.setSections(sections);
log.info("Building Sections for {}", identifier);
sectionsBuilderService.buildSections(classificationDocument);
sectionsBuilderService.addImagesToSections(classificationDocument);
// sectionsBuilderService.buildSections(classificationDocument);
// sectionsBuilderService.addImagesToSections(classificationDocument);
return classificationDocument;
}
@@ -342,7 +359,9 @@ public class LayoutParsingPipeline {
private void increaseDocumentStatistics(ClassificationPage classificationPage, ClassificationDocument document) {
document.getFontSizeCounter().addAll(classificationPage.getFontSizeCounter().getCountPerValue());
if (!classificationPage.isLandscape()) {
document.getFontSizeCounter().addAll(classificationPage.getFontSizeCounter().getCountPerValue());
}
document.getFontCounter().addAll(classificationPage.getFontCounter().getCountPerValue());
document.getTextHeightCounter().addAll(classificationPage.getTextHeightCounter().getCountPerValue());
document.getFontStyleCounter().addAll(classificationPage.getFontStyleCounter().getCountPerValue());
@@ -10,6 +10,8 @@ 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;
@@ -20,7 +22,6 @@ 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;
@@ -78,14 +79,6 @@ 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) {
@@ -12,15 +12,16 @@ import com.knecon.fforesight.service.layoutparser.processor.model.table.CleanRul
import com.knecon.fforesight.service.layoutparser.processor.model.text.StringFrequencyCounter;
import lombok.Data;
import lombok.Getter;
import lombok.NonNull;
import lombok.RequiredArgsConstructor;
import org.apache.pdfbox.pdmodel.documentinterchange.markedcontent.PDMarkedContent;
@Data
@RequiredArgsConstructor
public class ClassificationPage {
@NonNull
@Getter
private List<AbstractPageBlock> textBlocks;
private List<ClassifiedImage> images = new ArrayList<>();
@@ -76,14 +76,4 @@ public class Cell extends Rectangle {
return this.getHeight() >= MIN_SIZE && this.getWidth() >= MIN_SIZE;
}
public boolean nearlyIntersects(Cell other) {
if (this.getHeight() <= 0 || other.getHeight() <= 0) {
return false;
}
double x0 = this.getX() + 2;
double y0 = this.getY() + 2;
return (other.x + other.width > x0 && other.y + other.height > y0 && other.x < x0 + this.getWidth() - 2 && other.y < y0 + this.getHeight() - 2);
}
}
@@ -1,12 +1,14 @@
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.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;
@@ -19,7 +21,7 @@ import lombok.extern.slf4j.Slf4j;
@Slf4j
public class TablePageBlock extends AbstractPageBlock {
private final TreeMap<CellPosition, Cell> cellTreeMap = new TreeMap<>();
private final TreeMap<CellPosition, Cell> cells = new TreeMap<>();
private final int rotation;
@Getter
@@ -28,14 +30,10 @@ 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();
@@ -52,7 +50,6 @@ public class TablePageBlock extends AbstractPageBlock {
return getColCount() == 0 || getRowCount() == 0;
}
public List<List<Cell>> getRows() {
if (rows == null) {
@@ -83,10 +80,7 @@ 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);
}
@@ -126,8 +120,7 @@ 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);
}
@@ -142,8 +135,7 @@ 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);
}
}
@@ -159,7 +151,7 @@ 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 = cellTreeMap.get(new CellPosition(j, i));
Cell cell = cells.get(new CellPosition(j, i));
if (cell != null) {
lastRow.add(cell);
}
@@ -170,7 +162,7 @@ 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 = cellTreeMap.get(new CellPosition(j, i));
Cell cell = cells.get(new CellPosition(j, i));
if (cell != null) {
lastRow.add(cell);
}
@@ -181,7 +173,7 @@ 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 = cellTreeMap.get(new CellPosition(i, j)); // JAVA_8 use getOrDefault()
Cell cell = cells.get(new CellPosition(i, j)); // JAVA_8 use getOrDefault()
if (cell != null) {
lastRow.add(cell);
}
@@ -195,6 +187,17 @@ 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()) {
@@ -203,12 +206,11 @@ public class TablePageBlock extends AbstractPageBlock {
cells.removeIf(cell -> cell.getWidth() < 1.1 || cell.getHeight() < 1.1);
List<List<Cell>> rowsOfCellsMatrix = calculateTableStructure(cells);
List<List<Cell>> rowsOfCells = calculateStructure(cells);
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);
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);
}
}
@@ -219,36 +221,29 @@ 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 as a rows of cells matrix
* @return TablePageBlock Structure
*/
private List<List<Cell>> calculateTableStructure(List<Cell> cells) {
private List<List<Cell>> calculateStructure(List<Cell> cells) {
List<List<Cell>> matrix = new ArrayList<>();
if (cells.isEmpty()) {
return new ArrayList<>();
return matrix;
}
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();
List<List<Cell>> rowsOfCells = new ArrayList<>();
var sortedUniqueX = uniqueX.stream().sorted().toList();
var sortedUniqueY = uniqueY.stream().sorted().toList();
Float prevY = null;
for (Float y : sortedUniqueY) {
List<Cell> row = new ArrayList<>();
@@ -259,82 +254,43 @@ public class TablePageBlock extends AbstractPageBlock {
if (prevY != null && prevX != null) {
var cell = new Cell(new Point2D.Float(prevX, prevY), new Point2D.Float(x, y));
var intersectionCell = cells.stream().filter(c -> intersects(cell, c)).findFirst();
intersectionCell.ifPresent(value -> cell.getTextBlocks().addAll(value.getTextBlocks()));
if (cell.hasMinimumSize()) {
cells.stream()
.filter(cell::nearlyIntersects)
.forEach(intersectingCell -> cell.getTextBlocks().addAll(intersectingCell.getTextBlocks()));
row.add(cell);
}
}
prevX = x;
}
// 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);
if (prevY != null && prevX != null && !row.isEmpty()) {
matrix.add(row);
}
prevY = y;
}
Collections.reverse(rowsOfCells);
Collections.reverse(matrix);
// 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<>());
}
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;
return matrix;
}
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);
public boolean intersects(Cell cell1, Cell cell2) {
if (cell1.getHeight() <= 0 || cell2.getHeight() <= 0) {
return false;
}
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);
}
@Override
public String getText() {
@@ -358,7 +314,7 @@ 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;
}
}
@@ -372,6 +328,8 @@ public class TablePageBlock extends AbstractPageBlock {
}
public String getTextAsHtml() {
StringBuilder sb = new StringBuilder();
@@ -45,6 +45,9 @@ public class RedTextPosition {
@JsonIgnore
private String fontName;
@JsonIgnore
private RedTextPosition parent;
@SneakyThrows
public static RedTextPosition fromTextPosition(TextPosition textPosition) {
@@ -17,6 +17,7 @@ import lombok.AllArgsConstructor;
import lombok.Builder;
import lombok.Data;
import lombok.EqualsAndHashCode;
import lombok.Getter;
import lombok.NoArgsConstructor;
@EqualsAndHashCode(callSuper = true)
@@ -27,6 +28,7 @@ import lombok.NoArgsConstructor;
public class TextPageBlock extends AbstractPageBlock {
@Builder.Default
@Getter
private List<TextPositionSequence> sequences = new ArrayList<>();
@JsonIgnore
@@ -73,7 +75,7 @@ public class TextPageBlock extends AbstractPageBlock {
return sequences.get(0).getPageWidth();
}
public static TextPageBlock merge(List<TextPageBlock> textBlocksToMerge) {
@@ -82,6 +84,7 @@ public class TextPageBlock extends AbstractPageBlock {
return fromTextPositionSequences(sequences);
}
public static TextPageBlock fromTextPositionSequences(List<TextPositionSequence> wordBlockList) {
TextPageBlock textBlock = null;
@@ -133,7 +136,6 @@ public class TextPageBlock extends AbstractPageBlock {
}
/**
* Returns the minX value in pdf coordinate system.
* Note: This needs to use Pdf Coordinate System where {0,0} rotated with the page rotation.
@@ -55,6 +55,18 @@ public class TextPositionSequence implements CharSequence {
}
public TextPositionSequence(List<RedTextPosition> textPositions, int page) {
this.textPositions = textPositions;
this.page = page;
this.dir = TextDirection.fromDegrees(textPositions.get(0).getDir());
this.rotation = textPositions.get(0).getRotation();
this.pageHeight = textPositions.get(0).getPageHeight();
this.pageWidth = textPositions.get(0).getPageWidth();
this.isParagraphStart = false;
}
@Override
public int length() {
@@ -1,83 +0,0 @@
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<>();
visualLayoutParsingResponse.getData().forEach(tableData -> signatures.computeIfAbsent(tableData.getPage_idx(), 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,false,pageNumber);
signatures.add(signature);
}
});
return signatures;
}
}
@@ -1,18 +0,0 @@
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;
}
@@ -1,19 +0,0 @@
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;
}
@@ -1,21 +0,0 @@
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;
}
@@ -1,23 +0,0 @@
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;
}
@@ -1,22 +0,0 @@
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;
}
@@ -25,8 +25,7 @@ import lombok.extern.slf4j.Slf4j;
@RequiredArgsConstructor
public class RulingCleaningService {
private static final float THRESHOLD_Y = 6;
private static final float THRESHOLD_X = 2;
private static final float THRESHOLD = 6;
public CleanRulings getCleanRulings(List<TableCells> tableCells, List<Ruling> rulings) {
@@ -82,7 +81,7 @@ public class RulingCleaningService {
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_X) {
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)));
@@ -109,7 +108,7 @@ public class RulingCleaningService {
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_Y) {
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)));
@@ -240,7 +240,7 @@ public class SectionsBuilderService {
}
private ClassificationSection buildTextBlock(List<AbstractPageBlock> wordBlockList, String lastHeadline) {
public ClassificationSection buildTextBlock(List<AbstractPageBlock> wordBlockList, String lastHeadline) {
ClassificationSection section = new ClassificationSection();
@@ -5,6 +5,7 @@ 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;
@@ -24,62 +25,55 @@ import com.knecon.fforesight.service.layoutparser.processor.utils.DoubleComparis
@Service
public class TableExtractionService {
private static final int MAX_TABLE_OUTER_POINT_TOLERANCE = 10;
private static final int MAX_TABLE_CONTAINED_CELLS_WITH_TEXT = 1;
private static final float SPREADSHEET_AREA_TOLERANCE = 0.001f;
private static final Comparator<Point2D> X_FIRST_POINT_COMPARATOR = (point1, point2) -> {
private static final Comparator<Point2D> X_FIRST_POINT_COMPARATOR = (arg0, arg1) -> {
int rv = 0;
float point1X = DoubleComparisons.round(point1.getX(), 2);
float point1Y = DoubleComparisons.round(point1.getY(), 2);
float point2X = DoubleComparisons.round(point2.getX(), 2);
float point2Y = DoubleComparisons.round(point2.getY(), 2);
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 (point1X > point2X) {
if (arg0X > arg1X) {
rv = 1;
} else if (point1X < point2X) {
} else if (arg0X < arg1X) {
rv = -1;
} else if (point1Y > point2Y) {
} else if (arg0Y > arg1Y) {
rv = 1;
} else if (point1Y < point2Y) {
} else if (arg0Y < arg1Y) {
rv = -1;
}
return rv;
};
private static final Comparator<Point2D> Y_FIRST_POINT_COMPARATOR = (point1, point2) -> {
private static final Comparator<Point2D> POINT_COMPARATOR = (arg0, arg1) -> {
int rv = 0;
float point1X = DoubleComparisons.round(point1.getX(), 2);
float point1Y = DoubleComparisons.round(point1.getY(), 2);
float point2X = DoubleComparisons.round(point2.getX(), 2);
float point2Y = DoubleComparisons.round(point2.getY(), 2);
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 (point1Y > point2Y) {
if (arg0Y > arg1Y) {
rv = 1;
} else if (point1Y < point2Y) {
} else if (arg0Y < arg1Y) {
rv = -1;
} else if (point1X > point2X) {
} else if (arg0X > arg1X) {
rv = 1;
} else if (point1X < point2X) {
} else if (arg0X < arg1X) {
rv = -1;
}
return rv;
};
private 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 boolean contains(Cell cell, double x, double y, double w, double h) {
private 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);
};
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);
}
/**
@@ -95,18 +89,22 @@ 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());
// 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);
List<TextPageBlock> toBeRemoved = new ArrayList<>();
for (AbstractPageBlock abstractPageBlock : page.getTextBlocks()) {
TextPageBlock textBlock = (TextPageBlock) abstractPageBlock;
for (Cell cell : cells) {
if (cell.hasMinimumSize() && doesCellContainTextBlock(cell, textBlock)) {
if (cell.hasMinimumSize() && contains(cell,
textBlock.getPdfMinX(),
textBlock.getPdfMinY(),
textBlock.getPdfMaxX() - textBlock.getPdfMinX(),
textBlock.getPdfMaxY() - textBlock.getPdfMinY())) {
cell.addTextBlock(textBlock);
toBeRemoved.add(textBlock);
break;
}
}
@@ -116,70 +114,39 @@ public class TableExtractionService {
DoubleComparisons.sort(cells, Rectangle.ILL_DEFINED_ORDER);
List<Rectangle> spreadsheetAreas = 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> containedCells = new ArrayList<>();
List<Cell> overlappingCells = new ArrayList<>();
for (Cell c : cells) {
if (c.hasMinimumSize() && area.contains(c)) {
containedCells.add(c);
if (c.hasMinimumSize() && c.intersects(area)) {
overlappingCells.add(c);
}
}
var containedCellsWithText = containedCells.stream()
.filter(cell -> !cell.getTextBlocks().isEmpty())
.count();
// verify if table would contain fewer cells with text than the threshold allows
if (containedCellsWithText >= MAX_TABLE_CONTAINED_CELLS_WITH_TEXT) {
tables.add(new TablePageBlock(containedCells, area, page.getRotation()));
cells.removeAll(containedCells);
}
tables.add(new TablePageBlock(overlappingCells, area, page.getRotation()));
}
for (TablePageBlock table : tables) {
int position = -1;
for (AbstractPageBlock pageBlock : page.getTextBlocks()) {
if (pageBlock instanceof TextPageBlock ? table.containsBlock((TextPageBlock) pageBlock) : table.contains(pageBlock) && position == -1) {
position = page.getTextBlocks().indexOf(pageBlock);
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);
}
}
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);
}
private boolean doesCellContainTextBlock(Cell cell, TextPageBlock textBlock) {
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;
}
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 List<Cell> findCells(List<Ruling> horizontalRulingLines, List<Ruling> verticalRulingLines) {
public List<Cell> findCells(List<Ruling> horizontalRulingLines, List<Ruling> verticalRulingLines) {
// Fix for 211.pdf
for (Ruling r : horizontalRulingLines) {
@@ -193,7 +160,7 @@ public class TableExtractionService {
List<Cell> cellsFound = new ArrayList<>();
Map<Point2D, Ruling[]> intersectionPoints = Ruling.findIntersections(horizontalRulingLines, verticalRulingLines);
List<Point2D> intersectionPointsList = new ArrayList<>(intersectionPoints.keySet());
intersectionPointsList.sort(Y_FIRST_POINT_COMPARATOR);
intersectionPointsList.sort(POINT_COMPARATOR);
for (int i = 0; i < intersectionPointsList.size(); i++) {
Point2D topLeft = intersectionPointsList.get(i);
@@ -219,14 +186,13 @@ public class TableExtractionService {
continue;
}
for (Point2D yPoint : yPoints) {
// is there a horizontal edge b/w topLeft and yPoint ?
// 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])) {
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;
}
@@ -248,6 +214,7 @@ public class TableExtractionService {
Set<Point2D> pointSet = new HashSet<>();
Map<Point2D, Point2D> edgesH = new HashMap<>();
Map<Point2D, Point2D> edgesV = new HashMap<>();
int i = 0;
for (Rectangle cell : cells) {
for (Point2D pt : cell.getPoints()) {
@@ -264,9 +231,8 @@ public class TableExtractionService {
pointsSortX.sort(X_FIRST_POINT_COMPARATOR);
// Y first sort
List<Point2D> pointsSortY = new ArrayList<>(pointSet);
pointsSortY.sort(Y_FIRST_POINT_COMPARATOR);
pointsSortY.sort(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)) {
@@ -291,8 +257,7 @@ public class TableExtractionService {
Point2D nextVertex;
while (!edgesH.isEmpty()) {
ArrayList<PolygonVertex> polygon = new ArrayList<>();
Point2D first = edgesH.keySet()
.iterator().next();
Point2D first = edgesH.keySet().iterator().next();
polygon.add(new PolygonVertex(first, Direction.HORIZONTAL));
edgesH.remove(first);
@@ -336,14 +301,7 @@ public class TableExtractionService {
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_TABLE_OUTER_POINT_TOLERANCE) {
rectangles.add(new Rectangle(top - SPREADSHEET_AREA_TOLERANCE,
left - SPREADSHEET_AREA_TOLERANCE,
right - left + 2 * SPREADSHEET_AREA_TOLERANCE,
bottom - top + 2 * SPREADSHEET_AREA_TOLERANCE));
}
rectangles.add(new Rectangle(top, left, right - left, bottom - top));
}
return rectangles;
@@ -57,7 +57,7 @@ public class RedactManagerBlockificationService {
boolean isSplitByRuling = isSplitByRuling(minX, minY, maxX, maxY, word, horizontalRulingLines, verticalRulingLines);
boolean splitByDir = prev != null && !prev.getDir().equals(word.getDir());
if (prev != null && (lineSeparation || startFromTop || splitByX || splitByDir || isSplitByRuling)) {
if (prev != null && (splitByDir || isSplitByRuling)) {
Orientation prevOrientation = null;
if (!chunkBlockList.isEmpty()) {
@@ -167,7 +167,7 @@ public class RedactManagerBlockificationService {
}
private TextPageBlock buildTextBlock(List<TextPositionSequence> wordBlockList, int indexOnPage) {
public TextPageBlock buildTextBlock(List<TextPositionSequence> wordBlockList, int indexOnPage) {
TextPageBlock textBlock = null;
@@ -0,0 +1,153 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.List;
import java.util.Set;
import java.util.stream.Collectors;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPositionSequence;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Character;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.DisjointSets;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Line;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Zone;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service.LineBuilderService;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service.NearestNeighbourService;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service.ReadingOrderService;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service.SpacingService;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service.ZoneBuilderService;
import lombok.RequiredArgsConstructor;
@Service
@RequiredArgsConstructor
public class DocstrumSegmentationService {
private static final double MAX_VERTICAL_MERGE_DIST = 0.5;
private final NearestNeighbourService nearestNeighbourService;
private final SpacingService spacingService;
private final LineBuilderService lineBuilderService;
private final ZoneBuilderService zoneBuilderService;
private final ReadingOrderService readingOrderService;
public List<Zone> segmentPage(List<TextPositionSequence> textPositions) {
var positions = textPositions.stream().map(TextPositionSequence::getTextPositions).flatMap(List::stream).toList();
var characters = positions.stream().map(Character::new).collect(Collectors.toList());
nearestNeighbourService.findNearestNeighbors(characters);
var characterSpacing = spacingService.computeCharacterSpacing(characters);
var lineSpacing = spacingService.computeLineSpacing(characters);
var lines = lineBuilderService.buildLines(characters, characterSpacing, lineSpacing);
var zones = zoneBuilderService.buildZones(lines, characterSpacing, lineSpacing);
zones = mergeLinesInZones(zones, characterSpacing, Double.NEGATIVE_INFINITY, 0.0, 0.0, lineSpacing * MAX_VERTICAL_MERGE_DIST);
return readingOrderService.resolve(zones, false);
}
// private List<Zone> mergeZones(List<Zone> zones, double tolerance) {
//
// List<BxBounds> bounds = new ArrayList<BxBounds>(zones.size());
// for (List<ComponentLine> zone : zones) {
// BxBoundsBuilder builder = new BxBoundsBuilder();
// for (ComponentLine line : zone) {
// for (Component component : line.getComponents()) {
// builder.expand(component.getChunk().getBounds());
// }
// }
// bounds.add(builder.getBounds());
// }
//
// List<List<ComponentLine>> outputZones = new ArrayList<List<ComponentLine>>();
// mainFor:
// for (int i = 0; i < zones.size(); i++) {
// for (int j = 0; j < zones.size(); j++) {
// if (i == j || bounds.get(j) == null || bounds.get(i) == null) {
// continue;
// }
// if (BxModelUtils.contains(bounds.get(j), bounds.get(i), tolerance)) {
// zones.get(j).addAll(zones.get(i));
// bounds.set(i, null);
// continue mainFor;
// }
// }
// outputZones.add(zones.get(i));
// }
// return outputZones;
// }
private List<Zone> mergeLinesInZones(List<Zone> zones,
double wordSpacing,
double minHorizontalDistance,
double maxHorizontalDistance,
double minVerticalDistance,
double maxVerticalDistance) {
List<Zone> outputZones = new ArrayList<>(zones.size());
for (Zone zone : zones) {
outputZones.add(mergeLinesInZone(zone, wordSpacing, minHorizontalDistance, maxHorizontalDistance, minVerticalDistance, maxVerticalDistance));
}
return outputZones;
}
private Zone mergeLinesInZone(Zone zone,
double wordSpacing,
double minHorizontalDistance,
double maxHorizontalDistance,
double minVerticalDistance,
double maxVerticalDistance) {
DisjointSets<Line> sets = new DisjointSets<>(zone.getLines());
for (int i = 0; i < zone.getLines().size(); i++) {
Line li = zone.getLines().get(i);
for (int j = i + 1; j < zone.getLines().size(); j++) {
Line lj = zone.getLines().get(j);
double hDist = li.horizontalDistance(lj);
double vDist = li.verticalDistance(lj);
if (minHorizontalDistance <= hDist && hDist <= maxHorizontalDistance && minVerticalDistance <= vDist && vDist <= maxVerticalDistance) {
sets.union(li, lj);
} else if (minVerticalDistance <= vDist && vDist <= maxVerticalDistance && Math.abs(hDist - Math.min(li.getLength(), lj.getLength())) < 0.1) {
boolean componentOverlap = false;
int overlappingCount = 0;
for (Character ci : li.getCharacters()) {
for (Character cj : lj.getCharacters()) {
double dist = ci.overlappingDistance(cj);
if (dist > 2) {
componentOverlap = true;
}
if (dist > 0) {
overlappingCount++;
}
}
}
if (!componentOverlap && overlappingCount <= 2) {
sets.union(li, lj);
}
}
}
}
List<Line> outputZone = new ArrayList<>();
for (Set<Line> group : sets) {
List<Character> components = new ArrayList<>();
for (Line line : group) {
components.addAll(line.getCharacters());
}
components.sort(Comparator.comparingDouble(Character::getX));
outputZone.add(new Line(components, wordSpacing));
}
return new Zone(outputZone);
}
}
@@ -0,0 +1,32 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model;
public class AngleFilter {
protected double lowerAngle;
protected double upperAngle;
public AngleFilter(double lowerAngle, double upperAngle) {
if (lowerAngle < -Math.PI / 2) {
lowerAngle += Math.PI;
}
if (upperAngle >= Math.PI / 2) {
upperAngle -= Math.PI;
}
this.lowerAngle = lowerAngle;
this.upperAngle = upperAngle;
}
public boolean matches(Neighbor neighbor) {
if (lowerAngle <= upperAngle) {
return lowerAngle <= neighbor.getAngle() && neighbor.getAngle() < upperAngle;
} else {
return lowerAngle <= neighbor.getAngle() || neighbor.getAngle() < upperAngle;
}
}
}
@@ -0,0 +1,48 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model;
import java.awt.geom.Rectangle2D;
import lombok.Data;
@Data
public abstract class BoundingBox {
private Rectangle2D bBox;
public double getX() {
return bBox.getX();
}
public double getY() {
return bBox.getY();
}
public double getWidth() {
return bBox.getWidth();
}
public double getHeight() {
return bBox.getHeight();
}
public double getArea() {
return (bBox.getHeight() * bBox.getWidth());
}
public boolean contains(Rectangle2D contained, double tolerance) {
return bBox.getX() <= contained.getX() + tolerance && bBox.getY() <= contained.getY() + tolerance && bBox.getX() + bBox.getWidth() >= contained.getX() + contained.getWidth() - tolerance && bBox.getY() + bBox.getHeight() >= contained.getY() + contained.getHeight() - tolerance;
}
}
@@ -0,0 +1,84 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import com.knecon.fforesight.service.layoutparser.processor.model.text.RedTextPosition;
import lombok.Data;
@Data
public class Character {
private final double x;
private final double y;
private final RedTextPosition textPosition;
private List<Neighbor> neighbors = new ArrayList<>();
public Character(RedTextPosition chunk) {
this.x = chunk.getXDirAdj() + chunk.getWidthDirAdj() / 2;
this.y = chunk.getYDirAdj() + chunk.getHeightDir() / 2;
this.textPosition = chunk;
}
public double getHeight() {
return textPosition.getHeightDir();
}
public double distance(Character character) {
double dx = getX() - character.getX();
double dy = getY() - character.getY();
return Math.sqrt(dx * dx + dy * dy);
}
public double horizontalDistance(Character character) {
return Math.abs(getX() - character.getX());
}
public double verticalDistance(Character character) {
return Math.abs(getY() - character.getY());
}
public double overlappingDistance(Character other) {
double[] xs = new double[4];
double s = Math.sin(-0), c = Math.cos(-0);
xs[0] = c * x - s * y;
xs[1] = c * (x + textPosition.getWidthDirAdj()) - s * (y + textPosition.getHeightDir());
xs[2] = c * other.x - s * other.y;
xs[3] = c * (other.x + other.textPosition.getWidthDirAdj()) - s * (other.y + other.textPosition.getHeightDir());
boolean overlapping = xs[1] >= xs[2] && xs[3] >= xs[0];
Arrays.sort(xs);
return Math.abs(xs[2] - xs[1]) * (overlapping ? 1 : -1);
}
public void setNeighbors(List<Neighbor> neighbors) {
this.neighbors = neighbors;
}
public double angle(Character character) {
if (getX() > character.getX()) {
return Math.atan2(getY() - character.getY(), getX() - character.getX());
} else {
return Math.atan2(character.getY() - getY(), character.getX() - getX());
}
}
}
@@ -0,0 +1,194 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model;
import java.util.AbstractSet;
import java.util.Collection;
import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;
import java.util.NoSuchElementException;
import java.util.Set;
public class DisjointSets<E> implements Iterable<Set<E>> {
private final Map<E, Entry<E>> map = new HashMap<>();
public DisjointSets(Collection<? extends E> collection) {
for (E element : collection) {
map.put(element, new Entry<E>(element));
}
}
public boolean areTogether(E e1, E e2) {
return map.get(e1).findRepresentative() == map.get(e2).findRepresentative();
}
public void union(E e1, E e2) {
Entry<E> r1 = map.get(e1).findRepresentative();
Entry<E> r2 = map.get(e2).findRepresentative();
if (r1 != r2) {
if (r1.size <= r2.size) {
r2.mergeWith(r1);
} else {
r1.mergeWith(r2);
}
}
}
@Override
public Iterator<Set<E>> iterator() {
return new Iterator<>() {
private final Iterator<Entry<E>> iterator = map.values().iterator();
private Entry<E> nextRepresentative;
{
findNextRepresentative();
}
@Override
public boolean hasNext() {
return nextRepresentative != null;
}
@Override
public Set<E> next() {
if (nextRepresentative == null) {
throw new NoSuchElementException();
}
Set<E> result = nextRepresentative.asSet();
findNextRepresentative();
return result;
}
private void findNextRepresentative() {
while (iterator.hasNext()) {
Entry<E> candidate = iterator.next();
if (candidate.isRepresentative()) {
nextRepresentative = candidate;
return;
}
}
nextRepresentative = null;
}
@Override
public void remove() {
throw new UnsupportedOperationException();
}
};
}
private static class Entry<E> {
private int size = 1;
private final E value;
private Entry<E> parent = this;
private Entry<E> next = null;
private Entry<E> last = this;
Entry(E value) {
this.value = value;
}
void mergeWith(Entry<E> otherRepresentative) {
size += otherRepresentative.size;
last.next = otherRepresentative;
last = otherRepresentative.last;
otherRepresentative.parent = this;
}
Entry<E> findRepresentative() {
Entry<E> representative = parent;
while (representative.parent != representative) {
representative = representative.parent;
}
for (Entry<E> entry = this; entry != representative; ) {
Entry<E> nextEntry = entry.parent;
entry.parent = representative;
entry = nextEntry;
}
return representative;
}
boolean isRepresentative() {
return parent == this;
}
Set<E> asSet() {
return new AbstractSet<E>() {
@Override
public Iterator<E> iterator() {
return new Iterator<E>() {
private Entry<E> nextEntry = findRepresentative();
@Override
public boolean hasNext() {
return nextEntry != null;
}
@Override
public E next() {
if (nextEntry == null) {
throw new NoSuchElementException();
}
E result = nextEntry.value;
nextEntry = nextEntry.next;
return result;
}
@Override
public void remove() {
throw new UnsupportedOperationException();
}
};
}
@Override
public int size() {
return findRepresentative().size;
}
};
}
}
}
@@ -0,0 +1,91 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model;
public class Histogram {
private static final double EPSILON = 1.0e-6;
private final double min;
private final double resolution;
private double[] frequencies;
public Histogram(double minValue, double maxValue, double resolution) {
this.min = minValue - EPSILON;
double delta = maxValue - minValue + 2 * EPSILON;
int size = Math.max(1, (int) Math.round((maxValue - minValue) / resolution));
this.resolution = delta / size;
this.frequencies = new double[size];
}
public void kernelSmooth(double[] kernel) {
double[] newFrequencies = new double[frequencies.length];
int shift = (kernel.length - 1) / 2;
for (int i = 0; i < kernel.length; i++) {
int jStart = Math.max(0, i - shift);
int jEnd = Math.min(frequencies.length, frequencies.length + i - shift);
for (int j = jStart; j < jEnd; j++) {
newFrequencies[j - i + shift] += kernel[i] * frequencies[j];
}
}
frequencies = newFrequencies;
}
public double[] createGaussianKernel(double length, double stdDeviation) {
int r = (int) Math.round(length / resolution) / 2;
stdDeviation /= resolution;
int size = 2 * r + 1;
double[] kernel = new double[size];
double sum = 0;
double b = 2 * stdDeviation * stdDeviation;
double a = 1 / Math.sqrt(Math.PI * b);
for (int i = 0; i < size; i++) {
kernel[i] = a * Math.exp(-(i - r) * (i - r) / b);
sum += kernel[i];
}
for (int i = 0; i < size; i++) {
kernel[i] /= sum;
}
return kernel;
}
public void gaussianSmooth(double windowLength, double stdDeviation) {
kernelSmooth(createGaussianKernel(windowLength, stdDeviation));
}
public void add(double value) {
frequencies[(int) ((value - min) / resolution)] += 1.0;
}
public int getSize() {
return frequencies.length;
}
public double getPeakValue() {
int peakIndex = 0;
for (int i = 1; i < frequencies.length; i++) {
if (frequencies[i] > frequencies[peakIndex]) {
peakIndex = i;
}
}
int peakEndIndex = peakIndex + 1;
final double EPS = 0.0001;
while (peakEndIndex < frequencies.length && Math.abs(frequencies[peakEndIndex] - frequencies[peakIndex]) < EPS) {
peakEndIndex++;
}
return ((double) peakIndex + peakEndIndex) / 2 * resolution + min;
}
}
@@ -0,0 +1,165 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model;
import java.awt.geom.Rectangle2D;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPositionSequence;
import lombok.Data;
@Data
public class Line extends BoundingBox {
private static final double WORD_DISTANCE_MULTIPLIER = 0.2;
private final double x0;
private final double y0;
private final double x1;
private final double y1;
private final double height;
private final List<Character> characters;
private final List<TextPositionSequence> words = new ArrayList<>();
public Line(List<Character> characters, double wordSpacing) {
this.characters = characters;
if (characters.size() >= 2) {
// linear regression
double sx = 0.0, sxx = 0.0, sxy = 0.0, sy = 0.0;
for (Character character : characters) {
sx += character.getX();
sxx += character.getX() * character.getX();
sxy += character.getX() * character.getY();
sy += character.getY();
}
double b = (characters.size() * sxy - sx * sy) / (characters.size() * sxx - sx * sx);
double a = (sy - b * sx) / characters.size();
this.x0 = characters.get(0).getX();
this.y0 = a + b * this.x0;
this.x1 = characters.get(characters.size() - 1).getX();
this.y1 = a + b * this.x1;
} else {
Character character = characters.get(0);
double dx = character.getTextPosition().getWidthDirAdj() / 3;
double dy = dx * Math.tan(0);
this.x0 = character.getX() - dx;
this.x1 = character.getX() + dx;
this.y0 = character.getY() - dy;
this.y1 = character.getY() + dy;
}
height = computeHeight();
computeWords(wordSpacing * WORD_DISTANCE_MULTIPLIER);
buildBBox();
}
public double getAngle() {
return Math.atan2(y1 - y0, x1 - x0);
}
public double getLength() {
return Math.sqrt((x0 - x1) * (x0 - x1) + (y0 - y1) * (y0 - y1));
}
private double computeHeight() {
double sum = 0.0;
for (Character component : characters) {
sum += component.getHeight();
}
return sum / characters.size();
}
public double angularDifference(Line j) {
double diff = Math.abs(getAngle() - j.getAngle());
if (diff <= Math.PI / 2) {
return diff;
} else {
return Math.PI - diff;
}
}
public double horizontalDistance(Line other) {
double[] xs = new double[4];
xs[0] = x0;
xs[1] = x1;
xs[2] = other.x0;
xs[3] = other.x1;
boolean overlapping = xs[1] >= xs[2] && xs[3] >= xs[0];
Arrays.sort(xs);
return Math.abs(xs[2] - xs[1]) * (overlapping ? 1 : -1);
}
public double verticalDistance(Line other) {
double ym = (y0 + y1) / 2;
double yn = (other.y0 + other.y1) / 2;
return Math.abs(ym - yn) / Math.sqrt(1);
}
private void computeWords(double wordSpacing) {
TextPositionSequence word = new TextPositionSequence();
Character previous = null;
for (Character current : characters) {
if (previous != null) {
double dist = current.getTextPosition().getXDirAdj() - previous.getTextPosition().getXDirAdj() - previous.getTextPosition().getWidthDirAdj();
if (dist > wordSpacing) {
words.add(word);
word = new TextPositionSequence();
}
}
word.getTextPositions().add(current.getTextPosition());
previous = current;
}
words.add(word);
}
private void buildBBox() {
double minX = Double.POSITIVE_INFINITY;
double minY = Double.POSITIVE_INFINITY;
double maxX = Double.NEGATIVE_INFINITY;
double maxY = Double.NEGATIVE_INFINITY;
for (Character character : characters) {
minX = Math.min(minX, character.getTextPosition().getXDirAdj());
minY = Math.min(minY, character.getTextPosition().getYDirAdj());
maxX = Math.max(maxX, character.getTextPosition().getXDirAdj() + character.getTextPosition().getWidthDirAdj());
maxY = Math.max(maxY, character.getTextPosition().getYDirAdj() + character.getTextPosition().getHeightDir());
}
this.setBBox(new Rectangle2D.Double(minX, minY, maxX - minX, maxY - minY));
}
public String toString() {
StringBuilder sb = new StringBuilder();
words.forEach(word -> sb.append(word.toString()).append(" "));
return sb.toString().trim();
}
}
@@ -0,0 +1,36 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model;
import lombok.Getter;
public class Neighbor {
@Getter
private final double distance;
@Getter
private final double angle;
private final Character originCharacter;
@Getter
private final Character character;
public Neighbor(Character neighbor, Character origin) {
this.distance = neighbor.distance(origin);
this.angle = neighbor.angle(origin);
this.character = neighbor;
this.originCharacter = origin;
}
public double getHorizontalDistance() {
return character.horizontalDistance(originCharacter);
}
public double getVerticalDistance() {
return character.verticalDistance(originCharacter);
}
}
@@ -0,0 +1,50 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model;
import java.awt.geom.Rectangle2D;
import java.util.Comparator;
import java.util.List;
import lombok.Data;
@Data
public class Zone extends BoundingBox {
private List<Line> lines;
public Zone(List<Line> lines) {
lines.sort(Comparator.comparingDouble(Line::getY));
this.lines = lines;
buildBBox();
}
public void buildBBox() {
double minX = Double.POSITIVE_INFINITY;
double minY = Double.POSITIVE_INFINITY;
double maxX = Double.NEGATIVE_INFINITY;
double maxY = Double.NEGATIVE_INFINITY;
for (Line line : lines) {
minX = Math.min(minX, line.getX());
minY = Math.min(minY, line.getY());
maxX = Math.max(maxX, line.getX() + line.getWidth());
maxY = Math.max(maxY, line.getY() + line.getHeight());
}
this.setBBox(new Rectangle2D.Double(minX, minY, maxX - minX, maxY - minY));
}
public String toString() {
StringBuilder sb = new StringBuilder();
lines.forEach(line -> sb.append(line.toString()).append("\n"));
return sb.toString().trim();
}
}
@@ -0,0 +1,50 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.List;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.AngleFilter;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Character;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.DisjointSets;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Line;
@Service
public class LineBuilderService {
private static final double CHARACTER_SPACING_DISTANCE_MULTIPLIER = 3.5;
private static final double MAX_VERTICAL_CHARACTER_DISTANCE = 0.67;
private static final double ANGLE_TOLERANCE = Math.PI / 6;
public List<Line> buildLines(List<Character> characters, double characterSpacing, double lineSpacing) {
double maxHorizontalDistance = characterSpacing * CHARACTER_SPACING_DISTANCE_MULTIPLIER;
double maxVerticalDistance = lineSpacing * MAX_VERTICAL_CHARACTER_DISTANCE;
DisjointSets<Character> sets = new DisjointSets<>(characters);
AngleFilter filter = new AngleFilter(-ANGLE_TOLERANCE, ANGLE_TOLERANCE);
characters.forEach(character -> {
character.getNeighbors().forEach(neighbor -> {
double x = neighbor.getHorizontalDistance() / maxHorizontalDistance;
double y = neighbor.getVerticalDistance() / maxVerticalDistance;
if (filter.matches(neighbor) && Math.pow(x, 2) + Math.pow(y, 2) <= 1) {
sets.union(character, neighbor.getCharacter());
}
});
});
List<Line> lines = new ArrayList<>();
sets.forEach(group -> {
List<Character> lineComponents = new ArrayList<>(group);
lineComponents.sort(Comparator.comparingDouble(Character::getX));
lines.add(new Line(lineComponents, characterSpacing));
});
return lines;
}
}
@@ -0,0 +1,78 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.List;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Character;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Neighbor;
@Service
public class NearestNeighbourService {
private static final int NUMBER_OF_NEIGHBOURS = 8;
private static final double STEP = 16.0;
public void findNearestNeighbors(List<Character> characters) {
if (characters.isEmpty()) {
return;
}
characters.sort(Comparator.comparingDouble(Character::getX));
int maxNeighborCount = NUMBER_OF_NEIGHBOURS;
if (characters.size() <= NUMBER_OF_NEIGHBOURS) {
maxNeighborCount = characters.size() - 1;
}
for (int i = 0; i < characters.size(); i++) {
List<Neighbor> candidates = new ArrayList<>();
int start = i;
int end = i + 1;
double distance = Double.POSITIVE_INFINITY;
for (double searchDistance = 0; searchDistance < distance; ) {
searchDistance += STEP;
boolean newCandidatesFound = false;
while (start > 0 && characters.get(i).getX() - characters.get(start - 1).getX() < searchDistance) {
start--;
candidates.add(new Neighbor(characters.get(start), characters.get(i)));
clearLeastDistant(candidates, maxNeighborCount);
newCandidatesFound = true;
}
while (end < characters.size() && characters.get(end).getX() - characters.get(i).getX() < searchDistance) {
candidates.add(new Neighbor(characters.get(end), characters.get(i)));
clearLeastDistant(candidates, maxNeighborCount);
end++;
newCandidatesFound = true;
}
if (newCandidatesFound && candidates.size() >= maxNeighborCount) {
distance = candidates.get(maxNeighborCount - 1).getDistance();
}
}
clearLeastDistant(candidates, maxNeighborCount);
characters.get(i).setNeighbors(new ArrayList<>(candidates));
}
}
private void clearLeastDistant(List<Neighbor> candidates, int maxNeighborCount) {
if (candidates.size() > maxNeighborCount) {
candidates.sort(Comparator.comparingDouble(Neighbor::getDistance));
candidates.remove(candidates.remove(candidates.size() - 1));
}
}
}
@@ -0,0 +1,99 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.List;
import java.util.ListIterator;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.BoundingBox;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Zone;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.utils.DoubleUtils;
@Service
public class ReadingOrderService {
private static final double THRESHOLD = 1;
public List<Zone> resolve(List<Zone> zones, boolean yxOrder) {
if (zones.isEmpty() || zones.size() == 1) {
return zones;
}
if (yxOrder) {
zones.sort(Comparator.comparing(BoundingBox::getY, (o1, o2) -> DoubleUtils.compareDouble(o1, o2, THRESHOLD))
.thenComparing(BoundingBox::getX, (o1, o2) -> DoubleUtils.compareDouble(o1, o2, THRESHOLD)));
return zones;
}
return resolveMultiColumnReadingOder(zones);
}
private List<Zone> resolveMultiColumnReadingOder(List<Zone> zones) {
// Simple reading order resolver for multi column page layout as described here : https://pub.towardsai.net/advanced-rag-02-unveiling-pdf-parsing-b84ae866344e
// TODO implement a more fancy reading order resolver see https://github.com/BobLd/DocumentLayoutAnalysis/blob/master/README.md#reading-order
double minX = Double.POSITIVE_INFINITY;
double maxX = Double.NEGATIVE_INFINITY;
for (Zone zone : zones) {
if (zone.getX() < minX) {
minX = zone.getX();
}
if (zone.getX() + zone.getWidth() > maxX) {
maxX = zone.getX() + zone.getWidth();
}
}
double midLineXCoordinate = (minX + maxX) / 2;
List<Zone> leftOf = new ArrayList<>();
List<Zone> rightOf = new ArrayList<>();
List<Zone> middle = new ArrayList<>();
for (Zone zone : zones) {
if (zone.getX() < midLineXCoordinate && zone.getX() + zone.getWidth() < midLineXCoordinate) {
leftOf.add(zone);
} else if (zone.getX() > midLineXCoordinate && zone.getX() + zone.getWidth() > midLineXCoordinate) {
rightOf.add(zone);
} else {
middle.add(zone);
}
}
leftOf.sort(Comparator.comparing(BoundingBox::getY, (o1, o2) -> DoubleUtils.compareDouble(o1, o2, THRESHOLD))
.thenComparing(BoundingBox::getX, (o1, o2) -> DoubleUtils.compareDouble(o1, o2, THRESHOLD)));
rightOf.sort(Comparator.comparing(BoundingBox::getY, (o1, o2) -> DoubleUtils.compareDouble(o1, o2, THRESHOLD))
.thenComparing(BoundingBox::getX, (o1, o2) -> DoubleUtils.compareDouble(o1, o2, THRESHOLD)));
middle.sort(Comparator.comparing(BoundingBox::getY, (o1, o2) -> DoubleUtils.compareDouble(o1, o2, THRESHOLD))
.thenComparing(BoundingBox::getX, (o1, o2) -> DoubleUtils.compareDouble(o1, o2, THRESHOLD)));
List<Zone> sortedZones = new ArrayList<>();
sortedZones.addAll(leftOf);
sortedZones.addAll(rightOf);
ListIterator<Zone> itty = middle.listIterator();
while (itty.hasNext()) {
Zone current = itty.next();
for (int i = 0; i < sortedZones.size(); i++) {
if (current.getY() < sortedZones.get(i).getY()) {
sortedZones.add(i, current);
itty.remove();
break;
}
}
}
sortedZones.addAll(middle);
return sortedZones;
}
}
@@ -0,0 +1,56 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service;
import java.util.List;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.AngleFilter;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Character;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Histogram;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Neighbor;
@Service
public class SpacingService {
private static final double SPACING_HISTOGRAM_RESOLUTION = 0.5;
private static final double SPACING_HISTOGRAM_SMOOTHING_LENGTH = 2.5;
private static final double SPACING_HIST_SMOOTHING_STANDARD_DEVIATION = 0.5;
private static final double ANGLE_TOLERANCE = Math.PI / 6;
public double computeCharacterSpacing(List<Character> components) {
return computeSpacing(components, 0);
}
public double computeLineSpacing(List<Character> components) {
return computeSpacing(components, Math.PI / 2);
}
private double computeSpacing(List<Character> components, double angle) {
double maxDistance = Double.NEGATIVE_INFINITY;
for (Character component : components) {
for (Neighbor neighbor : component.getNeighbors()) {
maxDistance = Math.max(maxDistance, neighbor.getDistance());
}
}
Histogram histogram = new Histogram(0, maxDistance, SPACING_HISTOGRAM_RESOLUTION);
AngleFilter filter = new AngleFilter(angle - ANGLE_TOLERANCE, angle + ANGLE_TOLERANCE);
for (Character component : components) {
for (Neighbor neighbor : component.getNeighbors()) {
if (filter.matches(neighbor)) {
histogram.add(neighbor.getDistance());
}
}
}
histogram.gaussianSmooth(SPACING_HISTOGRAM_SMOOTHING_LENGTH, SPACING_HIST_SMOOTHING_STANDARD_DEVIATION);
return histogram.getPeakValue();
}
}
@@ -0,0 +1,90 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.service;
import java.util.ArrayList;
import java.util.List;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.DisjointSets;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Line;
import com.knecon.fforesight.service.layoutparser.processor.services.docstrum.model.Zone;
@Service
public class ZoneBuilderService {
private static final double MIN_HORIZONTAL_DISTANCE_MULTIPLIER = -0.5;
private static final double MAX_VERTICAL_DISTANCE_MULTIPLIER = 1.2;
private static final double MIN_HORIZONTAL_MERGE_DISTANCE_MULTIPLIER = -3.0;
private static final double MAX_VERTICAL_MERGE_DISTANCE_MULTIPLIER = 0.5;
private static final double MIN_LINE_SIZE_SCALE = 0.9;
private static final double MAX_LINE_SIZE_SCALE = 2.5;
private static final double ANGLE_TOLERANCE = Math.PI / 6;
public static final int MAX_ZONES = 300;
public List<Zone> buildZones(List<Line> lines, double characterSpacing, double lineSpacing) {
double minHorizontalDistance = characterSpacing * MIN_HORIZONTAL_DISTANCE_MULTIPLIER;
double maxVerticalDistance = lineSpacing * MAX_VERTICAL_DISTANCE_MULTIPLIER;
double minHorizontalMergeDistance = characterSpacing * MIN_HORIZONTAL_MERGE_DISTANCE_MULTIPLIER;
double maxVerticalMergeDistance = lineSpacing * MAX_VERTICAL_MERGE_DISTANCE_MULTIPLIER;
DisjointSets<Line> sets = new DisjointSets<>(lines);
double meanHeight = calculateMeanHeight(lines);
lines.forEach(outerLine -> //
lines.forEach(innerLine -> {
double scale = Math.min(outerLine.getHeight(), innerLine.getHeight()) / meanHeight;
scale = Math.max(MIN_LINE_SIZE_SCALE, Math.min(scale, MAX_LINE_SIZE_SCALE));
if (!sets.areTogether(outerLine, innerLine) && outerLine.angularDifference(innerLine) <= ANGLE_TOLERANCE) {
double horizontalDistance = outerLine.horizontalDistance(innerLine) / scale;
double verticalDistance = outerLine.verticalDistance(innerLine) / scale;
if (minHorizontalDistance <= horizontalDistance && verticalDistance <= maxVerticalDistance //
|| minHorizontalMergeDistance <= horizontalDistance && verticalDistance <= maxVerticalMergeDistance) {
sets.union(outerLine, innerLine);
}
}
}));
List<Zone> zones = new ArrayList<>();
sets.forEach(group -> {
zones.add(new Zone(new ArrayList<>(group)));
});
if (zones.size() > MAX_ZONES) {
List<Line> oneZoneLines = new ArrayList<>();
for (Zone zone : zones) {
oneZoneLines.addAll(zone.getLines());
}
return List.of(new Zone(oneZoneLines));
}
return zones;
}
private double calculateMeanHeight(List<Line> lines) {
double meanHeight = 0.0;
double weights = 0.0;
for (Line line : lines) {
double weight = line.getLength();
meanHeight += line.getHeight() * weight;
weights += weight;
}
meanHeight /= weights;
return meanHeight;
}
}
@@ -0,0 +1,18 @@
package com.knecon.fforesight.service.layoutparser.processor.services.docstrum.utils;
public class DoubleUtils {
public static int compareDouble(double d1, double d2, double precision) {
if (Double.isNaN(d1) || Double.isNaN(d2)) {
return Double.compare(d1, d2);
}
if (precision == 0) {
precision = 1;
}
long i1 = Math.round(d1 / precision);
long i2 = Math.round(d2 / precision);
return Long.valueOf(i1).compareTo(i2);
}
}
@@ -0,0 +1,270 @@
package com.knecon.fforesight.service.layoutparser.processor.services.visualization;
import java.awt.Color;
import java.awt.geom.Point2D;
import java.awt.geom.Rectangle2D;
import java.io.FileOutputStream;
import java.io.IOException;
import java.util.List;
import java.util.Map;
import org.apache.pdfbox.Loader;
import org.apache.pdfbox.pdmodel.PDDocument;
import org.apache.pdfbox.pdmodel.PDPage;
import org.apache.pdfbox.pdmodel.PDPageContentStream;
import org.apache.pdfbox.pdmodel.font.PDType1Font;
import org.apache.pdfbox.pdmodel.font.Standard14Fonts;
import org.apache.pdfbox.util.Matrix;
import org.springframework.core.io.ClassPathResource;
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.nodes.Document;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Page;
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.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) 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),
PdfVisualisationUtility.Options.builder().stroke(true).build());
}
pdDocument.save(out);
}
}
public static void drawRectanglesPerPageNumberedByLine(String filename, List<List<List<Rectangle2D>>> rectanglesPerPage, String tmpFileName) 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++) {
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());
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());
}
}
pdDocument.save(out);
}
}
private static int countNumberOfDigits(int num) {
int final_num = num;
if (final_num == 0) {
return 1;
}
int count = 0;
for (; final_num != 0; final_num /= 10) {
count++;
}
return count;
}
public static void drawDocumentGraph(PDDocument document, Document documentGraph) {
documentGraph.getDocumentTree().allEntriesInOrder().forEach(entry -> drawNode(document, entry));
}
public static void drawNode(PDDocument document, DocumentTree.Entry entry) {
Options options = buildStandardOptionsForNodes(entry);
drawBBoxAndLabelAndNumberOnPage(document, entry, options);
}
public static void drawTextBlock(PDDocument document, TextBlock textBlock, Options options) {
textBlock.getAtomicTextBlocks().forEach(atb -> drawAtomicTextBlock(document, atb, options));
}
public static void drawAtomicTextBlock(PDDocument document, AtomicTextBlock atomicTextBlock, Options options) {
drawRectangle2DList(document, atomicTextBlock.getPage().getNumber(), atomicTextBlock.getPositions().stream().toList(), options);
}
@SneakyThrows
private static void drawText(String string, PDDocument document, Point2D location, Integer pageNumber, Options options, boolean rotate) {
var pdPage = document.getPage(pageNumber - 1);
var contentStream = new PDPageContentStream(document, pdPage, PDPageContentStream.AppendMode.APPEND, true);
contentStream.setNonStrokingColor(options.getStrokeColor());
contentStream.setLineWidth(options.getStrokeWidth());
contentStream.beginText();
if (rotate) {
contentStream.setTextMatrix(Matrix.getRotateInstance(Math.toRadians(15), (float) location.getX(), (float) location.getY()));
} else {
contentStream.newLineAtOffset((float) location.getX(), (float) location.getY());
}
contentStream.setFont(new PDType1Font(Standard14Fonts.FontName.HELVETICA), 10);
contentStream.showText(string);
contentStream.endText();
contentStream.close();
}
@SneakyThrows
public static void drawRectangle2DList(PDDocument document, int pageNumber, List<Rectangle2D> rectCollection, Options 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 {
var contentStream = new PDPageContentStream(document, pdPage, PDPageContentStream.AppendMode.APPEND, true);
contentStream.setStrokingColor(options.getStrokeColor());
contentStream.setNonStrokingColor(options.getFillColor());
contentStream.setLineWidth(options.getStrokeWidth());
for (var r : rectCollection) {
contentStream.addRect((float) r.getMinX(), (float) r.getMinY(), (float) r.getWidth(), (float) r.getHeight());
if (options.isStroke() && options.isFill()) {
contentStream.fillAndStroke();
} else if (options.isStroke()) {
contentStream.stroke();
} else if (options.isFill()) {
contentStream.fill();
}
}
contentStream.close();
}
@SneakyThrows
public static void drawRectanglesAndLinesPerPage(String filename, List<List<Rectangle2D>> list, List<List<Rectangle2D>> rectanglesPerPage, String tmpFileName) {
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.drawLine2DList(pdDocument,
// pageNumber,
// list.get(pageNumber - 1),
// PdfVisualisationUtility.Options.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());
}
pdDocument.save(out);
}
}
@SneakyThrows
public static void drawLinesPerPage(String filename, List<List<Ruling>> linesPerPage, String tmpFileName) {
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.drawLine2DList(pdDocument,
pageNumber,
linesPerPage.get(pageNumber - 1),
PdfVisualisationUtility.Options.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) {
return Options.builder().stroke(true).strokeColor(switch (entry.getType()) {
case DOCUMENT -> Color.LIGHT_GRAY;
case HEADER, FOOTER -> Color.GREEN;
case PARAGRAPH -> Color.BLUE;
case HEADLINE -> Color.RED;
case SECTION -> Color.BLACK;
case TABLE -> Color.ORANGE;
case TABLE_CELL -> Color.GRAY;
case IMAGE -> Color.MAGENTA;
}).build();
}
private static void drawBBoxAndLabelAndNumberOnPage(PDDocument document, DocumentTree.Entry entry, Options options) {
Map<Page, Rectangle2D> rectanglesPerPage = entry.getNode().getBBox();
for (Page page : rectanglesPerPage.keySet()) {
Rectangle2D rectangle2D = rectanglesPerPage.get(page);
if (entry.getType() == NodeType.SECTION) {
rectangle2D = RectangleTransformations.pad(rectangle2D, 10, 10);
}
drawRectangle2DList(document, page.getNumber(), List.of(rectangle2D), options);
drawText(buildString(entry),
document,
new Point2D.Double(rectangle2D.getMinX(), rectangle2D.getMaxY() + 2),
page.getNumber(),
options,
entry.getType() == NodeType.TABLE_CELL);
}
}
private static String buildString(DocumentTree.Entry entry) {
return entry.getNode().getNumberOnPage() + ": " + entry.getTreeId() + ": " + entry.getType();
}
}
@@ -30,6 +30,8 @@ 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();
@@ -27,7 +27,6 @@ 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.server.utils.AbstractTest;
@@ -51,7 +50,6 @@ public class BdrJsonBuildTest extends AbstractTest {
file,
new ImageServiceResponse(),
new TableServiceResponse(),
new VisualLayoutParsingResponse(),
file.toString()));
}
@@ -33,7 +33,6 @@ 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;
@@ -99,7 +98,6 @@ public class HeadlinesGoldStandardIntegrationTest {
pdfFileResource.getFile(),
new ImageServiceResponse(),
new TableServiceResponse(),
new VisualLayoutParsingResponse(),
filePath));
var foundHeadlines = documentGraph.streamAllSubNodes()
@@ -25,7 +25,9 @@ public class LayoutparserEnd2EndTest extends AbstractTest {
@SneakyThrows
public void testLayoutParserEndToEnd() {
prepareStorage("files/bdr/Wie weiter bei Kristeneinrichtungen.pdf");
String s = "(";
String s1 = ")";
prepareStorage("files/Minimal Examples/WrongOrderPage1.pdf");
LayoutParsingRequest layoutParsingRequest = buildDefaultLayoutParsingRequest(LayoutParsingType.REDACT_MANAGER);
LayoutParsingFinishedEvent finishedEvent = layoutParsingPipeline.parseLayoutAndSaveFilesToStorage(layoutParsingRequest);
Arrays.stream(finishedEvent.message().split("\n")).forEach(log::info);
@@ -16,7 +16,6 @@ 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;
@@ -59,7 +58,6 @@ public class DocumentGraphJsonWritingTest extends BuildDocumentTest {
filename.toFile(),
new ImageServiceResponse(),
new TableServiceResponse(),
new VisualLayoutParsingResponse(),
filename.toFile().toString()));
DocumentData documentData = DocumentDataMapper.toDocumentData(documentGraph);
@@ -12,7 +12,6 @@ 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.visualization.LayoutGridService;
import com.knecon.fforesight.service.layoutparser.server.utils.BuildDocumentTest;
@@ -26,7 +25,7 @@ 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/Plenarprotokoll 1 (keine Druchsache!) (1).pdf";
String tmpFileName = "/tmp/" + Path.of(fileName).getFileName() + "_VIEWER.pdf";
var documentFile = new ClassPathResource(fileName).getFile();
@@ -54,10 +53,10 @@ public class ViewerDocumentTest extends BuildDocumentTest {
var documentFile = new ClassPathResource(fileName).getFile();
var classificationDocument = layoutParsingPipeline.parseLayout(LayoutParsingType.DOCUMINE,
documentFile,
new ImageServiceResponse(),
tableResponse,
new VisualLayoutParsingResponse(),Path.of(fileName).getFileName().toFile().toString());
documentFile,
new ImageServiceResponse(),
tableResponse,
Path.of(fileName).getFileName().toFile().toString());
ViewerDocumentService viewerDocumentService = new ViewerDocumentService(null);
LayoutGridService layoutGridService = new LayoutGridService(viewerDocumentService);
Document document = DocumentGraphFactory.buildDocumentGraph(classificationDocument);
@@ -66,4 +65,3 @@ public class ViewerDocumentTest extends BuildDocumentTest {
}
}
@@ -29,9 +29,10 @@ 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;
@@ -49,6 +50,12 @@ public class PdfSegmentationServiceTest extends AbstractTest {
@Autowired
private RedactManagerClassificationService redactManagerClassificationService;
@Autowired
private CvTableParsingAdapter cvTableParsingAdapter;
@Autowired
private ImageServiceResponseAdapter imageServiceResponseAdapter;
@Autowired
private SectionsBuilderService sectionsBuilderService;
@@ -57,11 +64,10 @@ public class PdfSegmentationServiceTest extends AbstractTest {
public ClassificationDocument buildClassificationDocument(File originDocument, TableServiceResponse tableServiceResponse) {
ClassificationDocument classificationDocument = layoutParsingPipeline.parseLayout(LayoutParsingType.REDACT_MANAGER,
originDocument,
new ImageServiceResponse(),
tableServiceResponse,
new VisualLayoutParsingResponse(),
"document");
originDocument,
new ImageServiceResponse(),
tableServiceResponse,
"document");
redactManagerClassificationService.classifyDocument(classificationDocument);
@@ -81,11 +87,11 @@ public class PdfSegmentationServiceTest extends AbstractTest {
@Test
public void tablesToHtmlDebugger() throws IOException {
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/T5 VV-640252-Page16.pdf");
ClassPathResource pdfFileResource = new ClassPathResource("files/SinglePages/A20622A izRMS (CZ) fRR Part B9_Page185.pdf");
ClassificationDocument document = buildClassificationDocument(pdfFileResource.getFile());
toHtml(document, "/tmp/T5.html");
toHtml(document, "/tmp/A20622A izRMS (CZ) fRR Part B9_Page185.html");
}
@@ -103,7 +109,6 @@ public class PdfSegmentationServiceTest extends AbstractTest {
}
@Disabled
@Test
public void testScanRotationBorderIsIgnored() throws IOException {
@@ -112,16 +117,8 @@ 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.
@@ -143,12 +140,12 @@ 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");
}
@@ -160,22 +157,11 @@ 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())
.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()).collect(Collectors.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);
}
@@ -185,37 +171,15 @@ 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())
.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()).collect(Collectors.toList()).get(0);
assertThat(firstTable.getColCount()).isEqualTo(8);
assertThat(firstTable.getRowCount()).isEqualTo(1);
TablePageBlock secondTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(1);
TablePageBlock secondTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.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();
}
@@ -225,37 +189,15 @@ 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())
.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()).collect(Collectors.toList()).get(0);
assertThat(firstTable.getColCount()).isEqualTo(9);
assertThat(firstTable.getRowCount()).isEqualTo(5);
TablePageBlock secondTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(1);
TablePageBlock secondTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.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();
}
@@ -265,41 +207,19 @@ 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())
.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()).collect(Collectors.toList()).get(0);
assertThat(firstTable.getColCount()).isEqualTo(8);
assertThat(firstTable.getRowCount()).isEqualTo(1);
TablePageBlock secondTable = document.getSections()
.stream()
.flatMap(paragraph -> paragraph.getTables()
.stream())
.toList()
.get(1);
TablePageBlock secondTable = document.getSections().stream().flatMap(paragraph -> paragraph.getTables().stream()).collect(Collectors.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
@Test // Non-sense 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");
@@ -310,8 +230,8 @@ public class PdfSegmentationServiceTest extends AbstractTest {
validateTable(document, 0, 1, 1, 0, 0);
validateTable(document, 1, 2, 2, 0, 0);
validateTable(document, 2, 4, 19, 12, 0);
validateTable(document, 3, 2, 12, 0, 0);
validateTable(document, 2, 6, 20, 0, 0);
validateTable(document, 3, 7, 31, 0, 0);
}
@@ -345,30 +265,29 @@ 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);
@@ -660,109 +579,10 @@ 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, 5, 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, 6);
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);
validateTable(document, 5, 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;
@@ -783,19 +603,9 @@ 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().isEmpty())
.toList().size();
int emptyCellsFoundFound = rows.stream().flatMap(List::stream).toList().stream().filter(f -> f.toString().equals("")).toList().size();
for (List<Cell> row : table.getRows()) {
row.forEach(r -> System.out.println(r.toString()));
@@ -810,20 +620,11 @@ 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);
@@ -836,11 +637,7 @@ 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);
}
@@ -21,14 +21,15 @@ 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.server.utils.BuildDocumentTest;
import com.knecon.fforesight.service.layoutparser.server.utils.visualizations.PdfDraw;
import com.knecon.fforesight.service.viewerdoc.service.ViewerDocumentService;
import lombok.SneakyThrows;
@@ -56,6 +57,9 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
@SneakyThrows
public void testTableExtraction() {
ViewerDocumentService viewerDocumentService = new ViewerDocumentService(null);
LayoutGridService layoutGridService = new LayoutGridService(viewerDocumentService);
ClassPathResource resource = new ClassPathResource("files");
List<String> pdfFileNames = Files.walk(resource.getFile().toPath())
.filter(path -> path.getFileName().toString().endsWith(".pdf"))
@@ -63,8 +67,8 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
.map(Path::toString)
.toList();
for (String pdfFileName : pdfFileNames) {
writeJsons(Path.of(pdfFileName));
for (int i = 0; i < pdfFileNames.size(); i++) {
writeJsons(Path.of(pdfFileNames.get(i)));
}
}
@@ -76,23 +80,21 @@ 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())) {
String tmpFileNameBefore = "/tmp/before." + filename.getFileName().toString();
if (!compareStructures(documentDataBefore.getDocumentStructure(), documentDataAfter.getDocumentStructure(), filename.getFileName().toString())) {
String tmpFileNameBefore = "C:/Users/YANNIK~1/AppData/Local/Temp/before." + filename.getFileName().toString();
try (PDDocument pdDocument = Loader.loadPDF(filename.toFile())) {
PdfDraw.drawDocumentGraph(pdDocument, documentGraphBefore);
pdDocument.save(tmpFileNameBefore);
}
String tmpFileNameAfter = "/tmp/after." + filename.getFileName().toString();
String tmpFileNameAfter = "C:/Users/YANNIK~1/AppData/Local/Temp/after." + filename.getFileName().toString();
try (PDDocument pdDocument = Loader.loadPDF(filename.toFile())) {
PdfDraw.drawDocumentGraph(pdDocument, documentGraphAfter);
pdDocument.save(tmpFileNameAfter);
@@ -103,9 +105,9 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
@SneakyThrows
private boolean compareStructures(DocumentStructure structure1, DocumentStructure structure2) {
private boolean compareStructures(DocumentStructure structure1, DocumentStructure structure2, String pdfName) {
List<Table> listStructure1 = structure1.streamAllEntries()
List listStructure1 = structure1.streamAllEntries()
.filter(entryData -> entryData.getType().equals(NodeType.TABLE))
.map(DocumentStructure.EntryData::getProperties)
.map(properties -> {
@@ -115,7 +117,7 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
})
.toList();
List<Table> listStructure2 = structure2.streamAllEntries()
List listStructure2 = structure2.streamAllEntries()
.filter(entryData -> entryData.getType().equals(NodeType.TABLE))
.map(DocumentStructure.EntryData::getProperties)
.map(properties -> {
@@ -126,8 +128,8 @@ public class RulingCleaningServiceTest extends BuildDocumentTest {
.toList();
for (int i = 0; i < listStructure1.size(); i++) {
Table tableNode1 = listStructure1.get(i);
Table tableNode2 = listStructure2.get(i);
Table tableNode1 = (Table) listStructure1.get(i);
Table tableNode2 = (Table) listStructure2.get(i);
if (tableNode1.getNumberOfRows() != tableNode2.getNumberOfRows() || tableNode1.getNumberOfCols() != tableNode2.getNumberOfCols()) {
return false;
}
@@ -20,7 +20,6 @@ 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;
@@ -52,8 +51,6 @@ 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";
@@ -99,7 +96,7 @@ public abstract class AbstractTest {
@SneakyThrows
protected LayoutParsingRequest prepareStorage(String file) {
return prepareStorage(file, "cv_table_parsing_response/empty.json", "image_service_response/empty.json","visual_layout_parsing_response/empty.json");
return prepareStorage(file, "cv_table_parsing_response/empty.json", "image_service_response/empty.json");
}
@@ -119,7 +116,6 @@ 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)
@@ -140,16 +136,6 @@ 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) {
@@ -161,17 +147,6 @@ 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() {
@@ -10,7 +10,6 @@ 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;
@@ -26,7 +25,7 @@ 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(), new VisualLayoutParsingResponse(),filename);
return layoutParsingPipeline.parseLayout(layoutParsingType, fileResource, layoutParsingStorageService.getImagesFile(IMAGE_FILE_ID), new TableServiceResponse(), filename);
}
@@ -1,8 +0,0 @@
{
"dossierId": "123",
"fileId": "123",
"targetFileExtension": "ORIGIN.pdf.gz",
"responseFileExtension": "EXTRACTED_TABLES.json.gz",
"data": [],
"X_TENANT_ID": ""
}