Marmelator/layout-parser
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 289 Wiki Activity

RED-9974: wip

Browse Source
This commit is contained in:
Kilian Schuettler 2024-09-13 12:36:59 +02:00 committed by Dominique Eifländer
parent 31bf4ba8c8
commit 1337c56591
8 changed files with 166 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/model/text/TextPageBlock.java
View File

@ -40,6 +40,8 @@ public class TextPageBlock extends AbstractPageBlock {
private double mostPopularWordSpaceWidth;
private boolean underlined;
private double highestFontSize;
private PageBlockType classification;
@ -140,6 +142,9 @@ public class TextPageBlock extends AbstractPageBlock {
setMostPopularWordHeight(lineHeightFrequencyCounter.getMostPopular());
setMostPopularWordSpaceWidth(spaceFrequencyCounter.getMostPopular());
setHighestFontSize(fontSizeFrequencyCounter.getHighest());
setUnderlined(sequences.stream()
.allMatch(TextPositionSequence::isUnderline));
}
@ -199,19 +204,7 @@ public class TextPageBlock extends AbstractPageBlock {
@Override
public String toString() {
StringBuilder builder = new StringBuilder();
for (int i = 0; i < sequences.size(); i++) {
String sequenceAsString = sequences.get(i).toString();
// Fix for missing Whitespace. This is recognized in getSequences method. See PDFTextStripper Line 1730.
if (i != 0 && sequences.get(i - 1).charAt(sequences.get(i - 1).length() - 1) != ' ' && sequenceAsString.charAt(0) != ' ') {
builder.append(' ');
}
builder.append(sequenceAsString);
}
return builder.toString();
return getText();
}

1
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/blockification/RedactManagerBlockificationService.java
View File

@ -161,7 +161,6 @@ public class RedactManagerBlockificationService {
}
if (!textPositions.isEmpty()) {
visualizations.addTextBlockVisualizations(chunkBlockList.stream()
.map(tb -> (TextPageBlock) tb)
.toList(), textPositions.get(0).getPage());
}

150
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/classification/DocuMineClassificationService.java
View File

@ -1,5 +1,6 @@
package com.knecon.fforesight.service.layoutparser.processor.services.classification;
import java.util.ArrayList;
import java.util.List;
import java.util.Locale;
import java.util.regex.Matcher;
@ -24,9 +25,17 @@ import lombok.extern.slf4j.Slf4j;
@RequiredArgsConstructor
public class DocuMineClassificationService {
private static final Pattern HEADLINE_WITH_IDENTIFER_PATTERN = Pattern.compile("^([1-9]\\d?\\.){1,3}\\d{1,2}\\.?\\s[0-9A-Za-z \\[\\]]{2,50}", Pattern.CASE_INSENSITIVE);
private static final Pattern AT_LEAST_3_PATTERN = Pattern.compile("\\p{L}{3,}", Pattern.CASE_INSENSITIVE);
private static final Pattern HEADLINE_PATTTERN_WITH_SLASHES = Pattern.compile("^(\\d{1,1}\\.){1,3}\\d{1,2}\\.?\\s[a-z]{1,2}\\/[a-z]{1,2}.*");
private static final Pattern HEADLINE_WITH_2_IDENTIFER_PATTERN = Pattern.compile("^([1-9]\\d?\\.){1,3}\\d{1,2}\\.?\\s[a-z][0-9a-z \\[\\]]{2,50}", Pattern.CASE_INSENSITIVE);
private static final Pattern HEADLINE_WITH_SINGLE_IDENTIFER_PATTERN = Pattern.compile("^([0-9]\\.)\\s[a-z][0-9a-z \\[\\]]{2,50}", Pattern.CASE_INSENSITIVE);
private static final Pattern AT_LEAST_3_CHARS_PATTERN = Pattern.compile("\\p{L}{3,}", Pattern.CASE_INSENSITIVE);
private static final Pattern HEADLINE_PATTERN_WITH_SLASHES = Pattern.compile("^(\\d{1,1}\\.){1,3}\\d{1,2}\\.?\\s[a-z]{1,2}\\/[a-z]{1,2}.*");
private static final Pattern AMOUNT_PATTERN = Pattern.compile("^\\s*\\d+(?:\\.\\d+)?\\s*(?:ml|l|g|kg|mg|cm|mm|km|m|lb|oz|ppm|%|f)\\b", Pattern.CASE_INSENSITIVE);
private static final Pattern TABLE_OR_FIGURE_PATTER = Pattern.compile(
"^\\s*(?:table|continued\\s+table|appendix|figure)\\s+(?:[xvi]+|[a-z0-9]{1,3}(?:\\.[0-9]{1,3})*(?:-[0-9]{1,3})?)\\b",
Pattern.CASE_INSENSITIVE);
public static final int SEPARATION_THRESHOLD = 10; // if the min distance between a textblock and all its surrounding blocks, the regexes can be more lenient.
public static final int SURROUNDING_BLOCKS_RADIUS = 3; // number of surrounding blocks before and after the current textblock to be tested
public void classifyDocument(ClassificationDocument document) {
@ -38,6 +47,7 @@ public class DocuMineClassificationService {
HeadlineClassificationService headlineClassificationService = new HeadlineClassificationService();
for (ClassificationPage page : document.getPages()) {
document.getLayoutDebugLayer().addTextBlockVisualizations(page.getTextBlocks(), page.getPageNumber());
classifyPage(headlineClassificationService, page, document, headlineFontSizes);
}
}
@ -48,16 +58,35 @@ public class DocuMineClassificationService {
ClassificationDocument document,
List<Double> headlineFontSizes) {
for (AbstractPageBlock textBlock : page.getTextBlocks()) {
List<AbstractPageBlock> textBlocks = page.getTextBlocks();
for (int i = 0; i < textBlocks.size(); i++) {
AbstractPageBlock textBlock = textBlocks.get(i);
if (textBlock instanceof TextPageBlock) {
classifyBlock(headlineClassificationService, (TextPageBlock) textBlock, page, document, headlineFontSizes);
List<AbstractPageBlock> surroundingBlocks = getSurroundingBlocks(i, textBlocks);
classifyBlock(headlineClassificationService, (TextPageBlock) textBlock, surroundingBlocks, page, document, headlineFontSizes);
}
}
}
private List<AbstractPageBlock> getSurroundingBlocks(int originalIndex, List<AbstractPageBlock> textBlocks) {
int start = Math.max(originalIndex - SURROUNDING_BLOCKS_RADIUS, 0);
int end = Math.min(originalIndex + SURROUNDING_BLOCKS_RADIUS, textBlocks.size());
List<AbstractPageBlock> surroundingBlocks = new ArrayList<>(2 * SURROUNDING_BLOCKS_RADIUS);
for (int i = start; i < end; i++) {
if (i == originalIndex) {
continue;
}
surroundingBlocks.add(textBlocks.get(i));
}
return surroundingBlocks;
}
private void classifyBlock(HeadlineClassificationService headlineClassificationService,
TextPageBlock textBlock,
List<AbstractPageBlock> surroundingBlocks,
ClassificationPage page,
ClassificationDocument document,
List<Double> headlineFontSizes) {
@ -65,9 +94,16 @@ public class DocuMineClassificationService {
log.debug("headlineFontSizes: {}", headlineFontSizes);
var bodyTextFrame = page.getBodyTextFrame();
Matcher headlineWithIdentifierMatcher = HEADLINE_WITH_IDENTIFER_PATTERN.matcher(textBlock.toString());
Matcher atLeast3Matcher = AT_LEAST_3_PATTERN.matcher(textBlock.toString());
Matcher headlineWithSlashesMatcher = HEADLINE_PATTTERN_WITH_SLASHES.matcher(textBlock.toString());
Matcher headlineWith2IdentifierMatcher = HEADLINE_WITH_2_IDENTIFER_PATTERN.matcher(textBlock.toString());
Matcher atLeast3Matcher = AT_LEAST_3_CHARS_PATTERN.matcher(textBlock.toString());
Matcher headlineWithSlashesMatcher = HEADLINE_PATTERN_WITH_SLASHES.matcher(textBlock.toString());
Matcher amountMatcher = AMOUNT_PATTERN.matcher(textBlock.toString());
Matcher tableOrFigureMatcher = TABLE_OR_FIGURE_PATTER.matcher(textBlock.toString());
Matcher headlineWithSingleIdentifierMatcher = HEADLINE_WITH_SINGLE_IDENTIFER_PATTERN.matcher(textBlock.toString());
boolean isAtLeast3Characters = atLeast3Matcher.reset().find();
boolean isTocItem = textBlock.getText().contains("..............");
boolean headlineWithSlashesMatches = headlineWithSlashesMatcher.reset().matches();
boolean isAmount = amountMatcher.reset().find();
if (textBlock.getClassification() != null && textBlock.getClassification().isHeadline()) {
headlineClassificationService.setLastHeadlineFromOutline(textBlock);
@ -103,54 +139,110 @@ public class DocuMineClassificationService {
&& (textBlock.getMostPopularWordHeight() > document.getTextHeightCounter().getMostPopular()
|| textBlock.getMostPopularWordFontSize() > document.getFontSizeCounter().getMostPopular())
&& PositionUtils.getApproxLineCount(textBlock) < 5.9
&& (textBlock.getMostPopularWordStyle().contains("bold")
&& Character.isDigit(textBlock.toString().charAt(0))
&& atLeast3Matcher.reset().find()
&& ((textBlock.getMostPopularWordStyle().contains("bold") || textBlock.isUnderlined())//
&& Character.isDigit(textBlock.toString().charAt(0)) //
&& isAtLeast3Characters //
&& !textBlock.toString().contains(":") //
|| textBlock.toString().equals(textBlock.toString().toUpperCase(Locale.ROOT))
&& atLeast3Matcher.reset().find()
&& !textBlock.toString().contains(":")
&& !textBlock.toString().startsWith("(")//
|| textBlock.toString().startsWith("APPENDIX") //
|| textBlock.toString().startsWith("FIGURE") //
|| textBlock.toString().startsWith("Continued TABLE") //
|| textBlock.toString().startsWith("TABLE"))
&& !textBlock.toString().endsWith(":")
&& atLeast3Matcher.reset().find()) {
&& isAtLeast3Characters
&& !isTocItem
&& !isAmount) {
PageBlockType headlineType = HeadlineClassificationService.headlineClassByFontSize(textBlock, headlineFontSizes);
headlineClassificationService.classifyHeadline(textBlock, headlineType);
document.setHeadlines(true);
} else if (headlineWithIdentifierMatcher.reset().find()
setAsHeadline(headlineClassificationService, textBlock, document, headlineFontSizes);
} else if (isAllCaps(textBlock) //
&& textBlock.getText().length() > 5 //
&& isAtLeast3Characters //
&& !isAmount//
&& !textBlock.toString().contains(":") //
&& !textBlock.toString().startsWith("(") //
&& PositionUtils.getApproxLineCount(textBlock) < 2.9) {
setAsHeadline(headlineClassificationService, textBlock, document, headlineFontSizes);
} else if (headlineWith2IdentifierMatcher.reset().find()
&& PositionUtils.getApproxLineCount(textBlock) < 2.9
&& atLeast3Matcher.reset().find()
&& !headlineWithSlashesMatcher.reset().matches()) {
&& isAtLeast3Characters
&& !headlineWithSlashesMatches
&& !isAmount
&& !isTocItem) {
PageBlockType headlineType = HeadlineClassificationService.headlineClassByFontSize(textBlock, headlineFontSizes);
headlineClassificationService.classifyHeadline(textBlock, headlineType);
document.setHeadlines(true);
setAsHeadline(headlineClassificationService, textBlock, document, headlineFontSizes);
} else if (!isTocItem //
&& hasSeparation(textBlock, surroundingBlocks) //
&& (textBlock.getMostPopularWordHeight() >= page.getTextHeightCounter().getMostPopular() //
|| textBlock.getMostPopularWordFontSize() >= page.getFontSizeCounter().getMostPopular()) //
&& PositionUtils.getApproxLineCount(textBlock) < 2.9 //
&& (tableOrFigureMatcher.reset().find() || headlineWithSingleIdentifierMatcher.reset().find()) //
&& !isAmount //
&& !headlineWithSlashesMatches) {
setAsHeadline(headlineClassificationService, textBlock, document, headlineFontSizes);
} else if (PositionUtils.isWithinBodyTextFrame(bodyTextFrame, textBlock)
&& textBlock.getMostPopularWordFontSize() == document.getFontSizeCounter().getMostPopular()
&& textBlock.getMostPopularWordStyle().equals("bold")
&& !document.getFontStyleCounter().getMostPopular().equals("bold")) {
textBlock.setClassification(PageBlockType.PARAGRAPH_BOLD);
} else if (PositionUtils.isWithinBodyTextFrame(bodyTextFrame, textBlock)
&& textBlock.getMostPopularWordFont().equals(document.getFontCounter().getMostPopular())
&& textBlock.getMostPopularWordStyle().equals(document.getFontStyleCounter().getMostPopular())
&& textBlock.getMostPopularWordFontSize() == document.getFontSizeCounter().getMostPopular()) {
textBlock.setClassification(PageBlockType.PARAGRAPH);
} else if (PositionUtils.isWithinBodyTextFrame(bodyTextFrame, textBlock)
&& textBlock.getMostPopularWordFontSize() == document.getFontSizeCounter().getMostPopular()
&& textBlock.getMostPopularWordStyle().equals("italic")
&& !document.getFontStyleCounter().getMostPopular().equals("italic")
&& PositionUtils.getApproxLineCount(textBlock) < 2.9) {
textBlock.setClassification(PageBlockType.PARAGRAPH_ITALIC);
} else if (PositionUtils.isWithinBodyTextFrame(bodyTextFrame, textBlock)) {
textBlock.setClassification(PageBlockType.PARAGRAPH);
} else {
textBlock.setClassification(PageBlockType.PARAGRAPH);
}
}
}
private static boolean isAllCaps(TextPageBlock textBlock) {
return textBlock.toString().equals(textBlock.toString().toUpperCase(Locale.ROOT));
}
private boolean hasSeparation(TextPageBlock textBlock, List<AbstractPageBlock> surroundingBlocks) {
return surroundingBlocks.stream()
.allMatch(surroundingBlock -> calculateSeparation(textBlock, surroundingBlock) > Math.pow(SEPARATION_THRESHOLD, 2));
}
private double calculateMinSeparation(TextPageBlock textBlock, List<AbstractPageBlock> surroundingBlocks) {
return surroundingBlocks.stream()
.mapToDouble(surroundingBlock -> calculateSeparation(textBlock, surroundingBlock))
.min().orElse(Double.MAX_VALUE);
}
private static double calculateSeparation(TextPageBlock textBlock, AbstractPageBlock surroundingBlock) {
return Math.pow(surroundingBlock.horizontalDistance(textBlock), 2) + Math.pow(surroundingBlock.verticalDistance(textBlock), 2);
}
private static void setAsHeadline(HeadlineClassificationService headlineClassificationService,
TextPageBlock textBlock,
ClassificationDocument document,
List<Double> headlineFontSizes) {
PageBlockType headlineType = HeadlineClassificationService.headlineClassByFontSize(textBlock, headlineFontSizes);
headlineClassificationService.classifyHeadline(textBlock, headlineType);
document.setHeadlines(true);
}
}

4
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/visualization/LayoutDebugLayer.java
View File

@ -18,10 +18,10 @@ import com.iqser.red.service.persistence.service.v1.api.shared.model.redactionlo
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.BoundingBox;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Line;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Zone;
import com.knecon.fforesight.service.layoutparser.processor.model.AbstractPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.outline.OutlineObject;
import com.knecon.fforesight.service.layoutparser.processor.model.table.CleanRulings;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Ruling;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPositionSequence;
import com.knecon.fforesight.service.layoutparser.processor.utils.CoordinateTransforms;
import com.knecon.fforesight.service.layoutparser.processor.utils.MarkedContentUtils;
@ -177,7 +177,7 @@ public class LayoutDebugLayer extends LayoutDebugLayerConfig {
}
public void addTextBlockVisualizations(List<TextPageBlock> textPageBlocks, int page) {
public void addTextBlockVisualizations(List<AbstractPageBlock> textPageBlocks, int page) {
if (!active) {
return;

3
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/visualization/LayoutGrid.java
View File

@ -88,6 +88,9 @@ public class LayoutGrid extends LayoutGridLayerConfig {
public void addTreeId(SemanticNode semanticNode) {
Page page = semanticNode.getFirstPage();
if (semanticNode.getBBox().get(page) == null) {
return;
}
addPlacedText(page, semanticNode.getBBox().get(page), semanticNode.getBBox().get(page), buildTreeIdString(semanticNode), 1, treeIds, TREEID_COLOR);
}

9
layoutparser-service/viewer-doc-processor/src/main/java/com/knecon/fforesight/service/viewerdoc/layers/OcrDebugLayerConfig.java
View File

@ -25,7 +25,7 @@ public class OcrDebugLayerConfig extends AbstractLayerGroup {
protected final Visualizations debugText = Visualizations.builder().layer(LayerIdentifier.KNECON_OCR_TEXT_DEBUG).visibleByDefault(true).build();
protected final Visualizations tableLines = Visualizations.builder().layer(LayerIdentifier.KNECON_OCR_LINE_DEBUG).visibleByDefault(true).build();
protected final Visualizations overlappedText = Visualizations.builder().layer(LayerIdentifier.KNECON_OCR_OVERLAPPED_TEXT).visibleByDefault(false).build();
protected final Visualizations overlappedText = Visualizations.builder().layer(LayerIdentifier.KNECON_OCR_OVERLAPPED_TEXT).visibleByDefault(true).build();
protected final Visualizations debugBBox = Visualizations.builder().layer(LayerIdentifier.KNECON_OCR_BBOX_DEBUG).visibleByDefault(false).build();
@ -35,4 +35,11 @@ public class OcrDebugLayerConfig extends AbstractLayerGroup {
return List.of(debugText, tableLines, debugBBox, overlappedText);
}
@Override
public boolean isVisibleByDefault() {
return true;
}
}

6
layoutparser-service/viewer-doc-processor/src/main/java/com/knecon/fforesight/service/viewerdoc/service/OutlineUtility.java
View File

@ -65,13 +65,13 @@ public class OutlineUtility {
@SneakyThrows
private static void deleteExistingOutline(PDFDoc doc) {
public static void deleteExistingOutline(PDFDoc doc) {
Bookmark firstBookmark = doc.getFirstBookmark();
while (firstBookmark != null && firstBookmark.isValid()) {
// while (firstBookmark != null && firstBookmark.isValid()) {
firstBookmark.delete();
firstBookmark = doc.getFirstBookmark();
}
// }
}

26
layoutparser-service/viewer-doc-processor/src/test/java/com/knecon/fforesight/service/viewerdoc/service/PageContentCleanerTest.java
View File

@ -4,6 +4,7 @@ import java.io.File;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.nio.file.Path;
import java.util.List;
import java.util.Set;
import org.junit.jupiter.api.AfterAll;
@ -12,6 +13,8 @@ import org.junit.jupiter.api.Disabled;
import org.junit.jupiter.api.Test;
import com.knecon.fforesight.service.viewerdoc.LayerIdentifier;
import com.knecon.fforesight.service.viewerdoc.layers.IdpLayerConfig;
import com.knecon.fforesight.service.viewerdoc.layers.OcrDebugLayerConfig;
import com.pdftron.pdf.ElementBuilder;
import com.pdftron.pdf.ElementReader;
import com.pdftron.pdf.ElementWriter;
@ -44,8 +47,8 @@ class PageContentCleanerTest {
@SneakyThrows
public void testContentCleaning() {
Path file = Path.of("/home/kschuettler/Downloads/ITEM 23_Absorção cutanea.pdf");
File tmpFile = new File("/tmp/ITEM 23_Absorção cutanea.pdf");
Path file = Path.of("/home/kschuettler/Downloads/pdf24_zusammengefügt.pdf");
File tmpFile = new File("/tmp/OCR_DEMO.pdf");
try (var in = new FileInputStream(file.toFile());//
var doc = new PDFDoc(in);//
var out = new FileOutputStream(tmpFile);//
@ -58,7 +61,12 @@ class PageContentCleanerTest {
.writer(pageWriter)
.reader(reader)
.elementBuilder(builder)
.markedContentToRemove(Set.of(LayerIdentifier.KNECON_LAYOUT.markedContentName()))
.markedContentToRemove(Set.of(LayerIdentifier.KNECON_OCR.markedContentName(),
LayerIdentifier.KNECON_AZURE_IDP.markedContentName(),
LayerIdentifier.KNECON_OCR_DEBUG.markedContentName(),
LayerIdentifier.IDP_TABLES.markedContentName(),
LayerIdentifier.IDP_KV_PAIRS.markedContentName(),
LayerIdentifier.IDP_SECTIONS.markedContentName()))
.build();
try (PageIterator iterator = doc.getPageIterator()) {
@ -74,4 +82,16 @@ class PageContentCleanerTest {
}
@Test
@SneakyThrows
public void activateLayersByDefault() {
Path file = Path.of("/tmp/OCR_TEST/pdf24_zusammengefügt (1).pdf/viewerDocument.pdf");
try (var in = new FileInputStream(file.toFile()); PDFDoc doc = new PDFDoc(in); var out = new FileOutputStream("/tmp/OCR_DEMO_OCRED.pdf")) {
PdftronLayerUtility.setOrderArrayForPresentGroups(doc, List.of(OcrDebugLayerConfig.CONFIG_INSTANCE, IdpLayerConfig.CONFIG_INSTANCE));
doc.save(out, SDFDoc.SaveMode.REMOVE_UNUSED, null);
}
}
}