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

Merge branch 'RED-10249' into 'main'

Browse Source 
RED-10249: regex found incorrectly due to wrong text sorting

See merge request fforesight/layout-parser!252
This commit is contained in:
Kilian Schüttler 2024-11-04 12:51:38 +01:00
commit f9b25c8157
14 changed files with 159 additions and 170 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

70
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/LayoutParsingPipeline.java
View File

@ -26,6 +26,8 @@ import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.No
import com.knecon.fforesight.service.layoutparser.internal.api.queue.LayoutParsingFinishedEvent;
import com.knecon.fforesight.service.layoutparser.internal.api.queue.LayoutParsingRequest;
import com.knecon.fforesight.service.layoutparser.internal.api.queue.LayoutParsingType;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Character;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Ruling;
import com.knecon.fforesight.service.layoutparser.processor.model.text.RedTextPosition;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextDirection;
import com.knecon.fforesight.service.layoutparser.processor.services.classification.ClassificationService;
@ -158,9 +160,9 @@ public class LayoutParsingPipeline {
}
if (!viewerDocumentFile.equals(originFile)) {
viewerDocumentFile.delete();
assert !viewerDocumentFile.exists() || viewerDocumentFile.delete();
}
originFile.delete();
assert !originFile.exists() || originFile.delete();
return LayoutParsingFinishedEvent.builder()
.identifier(layoutParsingRequest.identifier())
@ -240,12 +242,8 @@ public class LayoutParsingPipeline {
}
List<ClassificationPage> classificationPages = new ArrayList<>();
OutlineObject lastProcessedOutlineObject = null;
// parsing the structure elements could be useful as well
if (layoutParsingType != LayoutParsingType.REDACT_MANAGER_OLD) {
classificationDocument.setOutlineObjectTree(outlineExtractorService.getOutlineObjectTree(originDocument));
}
classificationDocument.setOutlineObjectTree(outlineExtractorService.getOutlineObjectTree(originDocument));
long pageCount = originDocument.getNumberOfPages();
@ -277,18 +275,15 @@ public class LayoutParsingPipeline {
if (layoutParsingType.equals(LayoutParsingType.DOCUMINE_OLD)) {
var lines = TextPositionOperations.groupByLine(new HashSet<>(words));
classificationDocument.getLayoutDebugLayer().addLineVisualizationsFromNestedTextPosition(lines, pageNumber);
words = TextPositionOperations.sortLines(lines);
words = TextPositionOperations.sortWords(lines);
}
classificationDocument.getLayoutDebugLayer().addTextVisualizations(words, pageNumber);
PDRectangle pdr = pdPage.getMediaBox();
int rotation = pdPage.getRotation();
boolean isLandscape = pdr.getWidth() > pdr.getHeight() && (rotation == 0 || rotation == 180) || pdr.getHeight() > pdr.getWidth() && (rotation == 90 || rotation == 270);
PDRectangle cropbox = pdPage.getCropBox();
classificationDocument.getLayoutDebugLayer().addRulingVisualization(stripper.getRulings(), pageNumber);
CleanRulings cleanRulings = rulingCleaningService.deduplicateAndStraightenRulings(pdfTableCells.get(pageNumber), stripper.getRulings());
List<Ruling> rulings = stripper.getRulings();
classificationDocument.getLayoutDebugLayer().addRulingVisualization(rulings, pageNumber);
CleanRulings cleanRulings = rulingCleaningService.deduplicateAndStraightenRulings(pdfTableCells.get(pageNumber), rulings);
PageInformation pageInformation = PageInformation.fromPDPage(pageNumber, pdPage);
List<Cell> emptyTableCells = TableExtractionService.findCells(cleanRulings.getHorizontals(), cleanRulings.getVerticals(), pageInformation);
@ -308,8 +303,7 @@ public class LayoutParsingPipeline {
.toList());
ClassificationPage classificationPage = switch (layoutParsingType) {
case REDACT_MANAGER_OLD ->
redactManagerBlockificationService.blockify(stripper.getWords(), cleanRulings, classificationDocument.getLayoutDebugLayer());
case REDACT_MANAGER_OLD -> redactManagerBlockificationService.blockify(stripper.getWords(), cleanRulings, classificationDocument.getLayoutDebugLayer());
case DOCUMINE_OLD -> docuMineBlockificationService.blockify(words, cleanRulings);
case DOCUMINE, REDACT_MANAGER, REDACT_MANAGER_PARAGRAPH_DEBUG, REDACT_MANAGER_WITHOUT_DUPLICATE_PARAGRAPH ->
docstrumBlockificationService.blockify(words, cleanRulings, true, classificationDocument.getLayoutDebugLayer(), layoutParsingType);
@ -317,27 +311,9 @@ public class LayoutParsingPipeline {
docstrumBlockificationService.blockify(words, cleanRulings, false, classificationDocument.getLayoutDebugLayer(), layoutParsingType);
};
classificationPage.setCleanRulings(cleanRulings);
classificationPage.setRotation(rotation);
classificationPage.setLandscape(isLandscape);
classificationPage.setPageNumber(pageNumber);
classificationPage.setPageWidth(cropbox.getWidth());
classificationPage.setPageHeight(cropbox.getHeight());
updateClassificationPage(pdPage, pdr, classificationPage, cleanRulings, pageNumber, pageInformation);
if (layoutParsingType != LayoutParsingType.REDACT_MANAGER_OLD) {
List<OutlineObject> outlineObjects = classificationDocument.getOutlineObjectTree().getOutlineObjectsPerPage().getOrDefault(pageNumber, new ArrayList<>());
OutlineObject notFoundOutlineObject = null;
if (lastProcessedOutlineObject != null && !lastProcessedOutlineObject.isFound()) {
lastProcessedOutlineObject.resetPoint();
notFoundOutlineObject = lastProcessedOutlineObject;
}
if (!outlineObjects.isEmpty()) {
classificationPage.setOutlineObjects(outlineObjects);
lastProcessedOutlineObject = blockificationPostprocessingService.sanitizeOutlineBlocks(classificationPage, notFoundOutlineObject);
}
classificationDocument.getLayoutDebugLayer().addOutlineObjects(outlineObjects, pageInformation);
}
blockificationPostprocessingService.findHeadlinesFromOutline(classificationDocument, pageNumber, classificationPage, pageInformation);
classificationDocument.getLayoutDebugLayer().addMarkedContentVisualizations(stripper.getMarkedContents(), pageNumber);
// MarkedContent needs to be converted at this point, otherwise it leads to GC Problems in Pdfbox.
@ -383,13 +359,31 @@ public class LayoutParsingPipeline {
}
private static void updateClassificationPage(PDPage pdPage,
PDRectangle pdr,
ClassificationPage classificationPage,
CleanRulings cleanRulings,
int pageNumber,
PageInformation pageInformation) {
int rotation = pdPage.getRotation();
boolean isLandscape = pdr.getWidth() > pdr.getHeight() && (rotation == 0 || rotation == 180) || pdr.getHeight() > pdr.getWidth() && (rotation == 90 || rotation == 270);
classificationPage.setCleanRulings(cleanRulings);
classificationPage.setRotation(rotation);
classificationPage.setLandscape(isLandscape);
classificationPage.setPageNumber(pageNumber);
classificationPage.setPageWidth((float) pageInformation.width());
classificationPage.setPageHeight((float) pageInformation.height());
}
private static void rotateDirAdjExactly(List<Word> words, PDPage pdPage) {
for (TextDirection dir : TextDirection.values()) {
double averageRotation = words.stream()
.map(Word::getTextPositions)
.map(Word::getCharacters)
.flatMap(Collection::stream)
.map(Character::getTextPosition)
.filter(pos -> pos.getDir().equals(dir))
.mapToDouble(RedTextPosition::getExactDir).average().orElse(0);

8
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/docstrum/DocstrumSegmentationService.java
View File

@ -80,16 +80,12 @@ public class DocstrumSegmentationService {
private List<Zone> computeZones(List<Word> textPositions, CleanRulings rulings, LayoutDebugLayer visualizations, TextDirection direction) {
List<RedTextPosition> positions = textPositions.stream()
List<Character> characters = textPositions.stream()
.filter(t -> t.getDir() == direction)
.map(Word::getTextPositions)
.map(Word::getCharacters)
.flatMap(List::stream)
.toList();
List<Character> characters = positions.stream()
.map(Character::new)
.collect(Collectors.toList());
nearestNeighbourService.findNearestNeighbors(characters);
double characterSpacing = spacingService.computeCharacterSpacing(characters);

42
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/docstrum/model/Line.java
View File

@ -36,18 +36,13 @@ public class Line extends TextBoundingBox {
@EqualsAndHashCode.Include
private final double y1;
private final double height;
private FontStyle fontStyle;
private final List<Character> characters;
private final List<Word> words = new ArrayList<>();
private final List<Word> words;
public Line(List<Character> characters, double wordSpacing) {
this.characters = characters;
if (characters.size() >= 2) {
// linear regression
double sx = 0.0;
@ -76,13 +71,25 @@ public class Line extends TextBoundingBox {
this.y0 = character.getY() - dy;
this.y1 = character.getY() + dy;
}
height = computeHeight();
computeWords(wordSpacing * WORD_DISTANCE_MULTIPLIER);
this.words = new ArrayList<>();
computeWords(characters, wordSpacing * WORD_DISTANCE_MULTIPLIER);
buildBBox();
computeFontStyle();
}
public Line(List<Word> words) {
this.words = words;
buildBBox();
x0 = getMinX();
y0 = getMinY();
x1 = getMaxX();
y1 = getMaxY();
computeFontStyle();
}
private void computeFontStyle() {
EnumMap<FontStyle, AtomicInteger> fontStyleCounter = new EnumMap<>(FontStyle.class);
@ -100,8 +107,7 @@ public class Line extends TextBoundingBox {
fontStyle = fontStyleCounter.entrySet()
.stream()
.max(Comparator.comparing(entry -> entry.getValue().get()))
.map(Map.Entry::getKey)
.orElse(FontStyle.REGULAR);
.map(Map.Entry::getKey).orElse(FontStyle.REGULAR);
}
@ -117,14 +123,6 @@ public class Line extends TextBoundingBox {
}
private double computeHeight() {
return characters.stream()
.map(Character::getHeight)
.reduce(0d, Double::sum) / characters.size();
}
public double angularDifference(Line j) {
double diff = Math.abs(getAngle() - j.getAngle());
@ -157,7 +155,7 @@ public class Line extends TextBoundingBox {
}
private void computeWords(double wordSpacing) {
private void computeWords(List<Character> characters, double wordSpacing) {
Word word = new Word();
Character previous = null;
@ -169,7 +167,7 @@ public class Line extends TextBoundingBox {
word = new Word();
}
}
word.getTextPositions().add(current.getTextPosition());
word.add(current);
previous = current;
}
words.add(word);
@ -178,9 +176,7 @@ public class Line extends TextBoundingBox {
private void buildBBox() {
this.setToBBoxOfComponents(characters.stream()
.map(Character::getTextPosition)
.toList());
this.setToBBoxOfComponents(words);
}

1
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/docstrum/model/Zone.java
View File

@ -18,7 +18,6 @@ public class Zone extends TextBoundingBox {
@SuppressWarnings("PMD.ConstructorCallsOverridableMethod")
public Zone(List<Line> lines) {
lines.sort(Comparator.comparingDouble(Line::getY0));
this.lines = lines;
setToBBoxOfComponents(lines);
}

73
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/docstrum/service/ZoneBuilderService.java
View File

@ -1,9 +1,7 @@
package com.knecon.fforesight.service.layoutparser.processor.docstrum.service;
import static com.knecon.fforesight.service.layoutparser.processor.model.SectionIdentifier.numericalIdentifierPattern;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.Collection;
import java.util.HashSet;
import java.util.List;
import java.util.Set;
@ -11,11 +9,12 @@ import java.util.stream.Collectors;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Character;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Line;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.UnionFind;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Zone;
import com.knecon.fforesight.service.layoutparser.processor.model.table.CleanRulings;
import com.knecon.fforesight.service.layoutparser.processor.model.text.Word;
import com.knecon.fforesight.service.layoutparser.processor.utils.TextPositionOperations;
@Service
public class ZoneBuilderService {
@ -31,7 +30,7 @@ public class ZoneBuilderService {
private static final double MAX_LINE_SIZE_SCALE = 2.5;
private static final double ANGLE_TOLERANCE = Math.PI / 6;
private static final double ANGLE_TOLERANCE = Math.toRadians(5);
private static final double MAX_VERTICAL_MERGE_DISTANCE = 0.5;
@ -114,64 +113,14 @@ public class ZoneBuilderService {
private Zone mergeLinesInZone(List<Line> lines, double characterSpacing, double lineSpacing) {
double maxHorizontalDistance = 0;
double minVerticalDistance = 0;
double maxVerticalDistance = lineSpacing * MAX_VERTICAL_MERGE_DISTANCE;
Set<Word> words = lines.stream()
.map(Line::getWords)
.flatMap(Collection::stream)
.collect(Collectors.toSet());
Collection<Set<Word>> groupedLines = TextPositionOperations.groupByLine(words);
UnionFind<Line> unionFind = new UnionFind<>(new HashSet<>(lines));
lines.forEach(outer -> {
lines.forEach(inner -> {
if (inner == outer) {
return;
}
double horizontalDistance = outer.horizontalDistance(inner);
double verticalDistance = outer.verticalDistance(inner);
if (horizontalDistance <= maxHorizontalDistance && minVerticalDistance <= verticalDistance && verticalDistance <= maxVerticalDistance) {
unionFind.union(outer, inner);
} else if (minVerticalDistance <= verticalDistance
&& verticalDistance <= maxVerticalDistance
&& Math.abs(horizontalDistance - Math.min(outer.getLength(), inner.getLength())) < 0.1) {
boolean characterOverlap = false;
int overlappingCount = 0;
for (Character outerCharacter : outer.getCharacters()) {
for (Character innerCharacter : inner.getCharacters()) {
double characterOverlapDistance = outerCharacter.overlappingDistance(innerCharacter);
if (characterOverlapDistance > 2) {
characterOverlap = true;
}
if (characterOverlapDistance > 0) {
overlappingCount++;
}
}
}
if (!characterOverlap && overlappingCount <= 2) {
unionFind.union(outer, inner);
}
}
});
});
List<Line> outputZone = new ArrayList<>();
for (Set<Line> group : unionFind.getGroups()) {
List<Character> characters = new ArrayList<>();
for (Line line : group) {
characters.addAll(line.getCharacters());
}
characters.sort(Comparator.comparingDouble(Character::getX));
outputZone.add(new Line(characters, characterSpacing));
}
return new Zone(outputZone.stream()
.sorted(Comparator.comparing(Line::getY0))
.collect(Collectors.toList()));
List<Line> sortedLines = TextPositionOperations.sortLines(groupedLines);
return new Zone(sortedLines);
}
}

53
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/model/text/Word.java
View File

@ -13,6 +13,7 @@ import java.util.stream.Collectors;
import org.apache.pdfbox.text.TextPosition;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Character;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.TextBoundingBox;
import lombok.AllArgsConstructor;
@ -38,8 +39,7 @@ public class Word extends TextBoundingBox implements CharSequence {
private int page;
@Builder.Default
private List<RedTextPosition> textPositions = new ArrayList<>();
private List<Character> characters = new ArrayList<>();
private boolean isParagraphStart;
private boolean strikethrough;
private boolean underline;
@ -49,8 +49,9 @@ public class Word extends TextBoundingBox implements CharSequence {
public Word(List<TextPosition> textPositions, int pageNumber, boolean isParagraphStart) {
this.textPositions = textPositions.stream()
this.characters = textPositions.stream()
.map(RedTextPosition::fromTextPosition)
.map(Character::new)
.collect(Collectors.toList());
this.page = pageNumber;
this.isParagraphStart = isParagraphStart;
@ -65,9 +66,9 @@ public class Word extends TextBoundingBox implements CharSequence {
}
public Word(List<RedTextPosition> textPositions, int page) {
public Word(List<Character> textPositions, int page) {
this.textPositions = textPositions;
this.characters = new ArrayList<>(textPositions);
this.page = page;
calculateBBoxAndHashcode();
}
@ -76,7 +77,7 @@ public class Word extends TextBoundingBox implements CharSequence {
@Override
public int length() {
return textPositions.size();
return characters.size();
}
@ -101,7 +102,7 @@ public class Word extends TextBoundingBox implements CharSequence {
public Word subSequence(int start, int end) {
var textPositionSequence = new Word();
textPositionSequence.textPositions = textPositions.subList(start, end);
textPositionSequence.characters = characters.subList(start, end);
textPositionSequence.page = page;
textPositionSequence.dir = dir;
textPositionSequence.setToBBoxOfComponents(getTextPositions());
@ -122,53 +123,59 @@ public class Word extends TextBoundingBox implements CharSequence {
public RedTextPosition textPositionAt(int index) {
return textPositions.get(index);
return characters.get(index).getTextPosition();
}
public void add(Word word, RedTextPosition textPosition) {
this.textPositions.add(textPosition);
this.characters.add(new Character(textPosition));
this.page = word.getPage();
calculateBBoxAndHashcode();
}
public void add(Character current) {
characters.add(current);
calculateBBoxAndHashcode();
}
public void add(TextPosition textPosition) {
this.textPositions.add(RedTextPosition.fromTextPosition(textPosition));
calculateBBoxAndHashcode();
add(new Character(RedTextPosition.fromTextPosition(textPosition)));
}
public double getTextHeightNoPadding() {
return textPositions.get(0).getHeightDirAdj();
return characters.get(0).getTextPosition().getHeightDirAdj();
}
public double getTextHeight() {
return textPositions.get(0).getHeightDirAdj() + HEIGHT_PADDING;
return characters.get(0).getTextPosition().getHeightDirAdj() + HEIGHT_PADDING;
}
public String getFont() {
if (textPositions.get(0).getFontName() == null) {
if (characters.get(0).getTextPosition().getFontName() == null) {
return "none";
}
return FONT_CLEANER.matcher(textPositions.get(0).getFontName().toLowerCase(Locale.ROOT)).replaceAll("");
return FONT_CLEANER.matcher(characters.get(0).getTextPosition().getFontName().toLowerCase(Locale.ROOT)).replaceAll("");
}
public String getFontStyle() {
if (textPositions.get(0).getFontName() == null) {
if (characters.get(0).getTextPosition().getFontName() == null) {
return STANDARD;
}
String lowercaseFontName = textPositions.get(0).getFontName().toLowerCase(Locale.ROOT);
String lowercaseFontName = characters.get(0).getTextPosition().getFontName().toLowerCase(Locale.ROOT);
if (lowercaseFontName.contains(BOLD) && lowercaseFontName.contains(ITALIC)) {
return BOLD_ITALIC;
@ -184,13 +191,13 @@ public class Word extends TextBoundingBox implements CharSequence {
public float getFontSize() {
return textPositions.get(0).getFontSizeInPt();
return characters.get(0).getTextPosition().getFontSizeInPt();
}
public float getSpaceWidth() {
return textPositions.get(0).getWidthOfSpace();
return characters.get(0).getTextPosition().getWidthOfSpace();
}
@ -244,6 +251,14 @@ public class Word extends TextBoundingBox implements CharSequence {
}
private List<RedTextPosition> getTextPositions() {
return characters.stream()
.map(Character::getTextPosition)
.toList();
}
public void transform(AffineTransform rotateInstance) {
for (RedTextPosition textPosition : getTextPositions()) {

26
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/blockification/BlockificationPostprocessingService.java
View File

@ -11,12 +11,14 @@ import org.apache.commons.text.similarity.LevenshteinDistance;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngineProto.LayoutEngine;
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.PageBlockType;
import com.knecon.fforesight.service.layoutparser.processor.model.SectionIdentifier;
import com.knecon.fforesight.service.layoutparser.processor.model.outline.OutlineObject;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.text.Word;
import com.knecon.fforesight.service.layoutparser.processor.utils.PageInformation;
import com.knecon.fforesight.service.layoutparser.processor.utils.TextNormalizationUtilities;
import lombok.Data;
@ -27,6 +29,24 @@ public class BlockificationPostprocessingService {
private static final float STRING_SIMILARITY_THRESHOLD = 0.1f;
public void findHeadlinesFromOutline(ClassificationDocument classificationDocument, int pageNumber, ClassificationPage classificationPage, PageInformation pageInformation) {
OutlineObject lastProcessedOutlineObject = null;
List<OutlineObject> outlineObjects = classificationDocument.getOutlineObjectTree().getOutlineObjectsPerPage().getOrDefault(pageNumber, new ArrayList<>());
OutlineObject notFoundOutlineObject = null;
if (lastProcessedOutlineObject != null && !lastProcessedOutlineObject.isFound()) {
lastProcessedOutlineObject.resetPoint();
notFoundOutlineObject = lastProcessedOutlineObject;
}
if (!outlineObjects.isEmpty()) {
classificationPage.setOutlineObjects(outlineObjects);
lastProcessedOutlineObject = sanitizeOutlineBlocks(classificationPage, notFoundOutlineObject);
}
classificationDocument.getLayoutDebugLayer().addOutlineObjects(outlineObjects, pageInformation);
}
public OutlineObject sanitizeOutlineBlocks(ClassificationPage classificationPage, OutlineObject notFoundOutlineObject) {
List<OutlineObject> outlineObjects = classificationPage.getOutlineObjects();
@ -329,8 +349,8 @@ public class BlockificationPostprocessingService {
if (index > 0) {
in = createSubSequence(sequence, 0, index);
} else if (endIndex < sequence.getTextPositions().size()) {
in = createSubSequence(sequence, endIndex, sequence.getTextPositions().size());
} else if (endIndex < sequence.length()) {
in = createSubSequence(sequence, endIndex, sequence.length());
}
return new SplitSequenceResult(in, out);
@ -339,7 +359,7 @@ public class BlockificationPostprocessingService {
private static Word createSubSequence(Word sequence, int start, int end) {
Word newSeq = new Word(new ArrayList<>(sequence.getTextPositions().subList(start, end)), sequence.getPage());
Word newSeq = new Word(new ArrayList<>(sequence.getCharacters().subList(start, end)), sequence.getPage());
newSeq.setParagraphStart(sequence.isParagraphStart());
return newSeq;
}

2
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/blockification/DocstrumBlockificationService.java
View File

@ -77,7 +77,7 @@ public class DocstrumBlockificationService {
.forEach(line -> {
line.getWords()
.forEach(word -> {
words.add(new Word(word.getTextPositions(), word.getPage()));
words.add(new Word(word.getCharacters(), word.getPage()));
});
});

4
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/classification/ClarifyndClassificationService.java
View File

@ -81,7 +81,7 @@ public class ClarifyndClassificationService {
&& (textBlock.getMostPopularWordStyle().equals("bold")
|| !document.getFontStyleCounter().getCountPerValue().containsKey("bold")
&& textBlock.getMostPopularWordFontSize() > document.getFontSizeCounter().getMostPopular() + 1)
&& textBlock.getWords().get(0).getTextPositions().get(0).getFontSizeInPt() >= textBlock.getMostPopularWordFontSize()) {
&& textBlock.getWords().get(0).getFontSize()>= textBlock.getMostPopularWordFontSize()) {
PageBlockType headlineType = HeadlineClassificationService.headlineClassByFontSize(textBlock, headlineFontSizes);
headlineClassificationService.classifyHeadline(textBlock, headlineType);
@ -91,7 +91,7 @@ public class ClarifyndClassificationService {
&& textBlock.getMostPopularWordStyle().equals("bold")
&& !document.getFontStyleCounter().getMostPopular().equals("bold")
&& PositionUtils.getApproxLineCount(textBlock) < 2.9
&& textBlock.getWords().get(0).getTextPositions().get(0).getFontSizeInPt() >= textBlock.getMostPopularWordFontSize()) {
&& textBlock.getWords().get(0).getFontSize() >= textBlock.getMostPopularWordFontSize()) {
PageBlockType headlineType = HeadlineClassificationService.headlineClassByFontSize(textBlock, headlineFontSizes);
headlineClassificationService.classifyHeadline(textBlock, headlineType);

4
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/classification/RedactManagerClassificationService.java
View File

@ -94,7 +94,7 @@ public class RedactManagerClassificationService {
&& (textBlock.getMostPopularWordStyle().equals("bold")
|| !document.getFontStyleCounter().getCountPerValue().containsKey("bold")
&& textBlock.getMostPopularWordFontSize() > document.getFontSizeCounter().getMostPopular() + 1)
&& textBlock.getWords().get(0).getTextPositions().get(0).getFontSizeInPt() >= textBlock.getMostPopularWordFontSize()) {
&& textBlock.getWords().get(0).getFontSize() >= textBlock.getMostPopularWordFontSize()) {
PageBlockType headlineType = HeadlineClassificationService.headlineClassByFontSize(textBlock, headlineFontSizes);
headlineClassificationService.classifyHeadline(textBlock, headlineType);
@ -104,7 +104,7 @@ public class RedactManagerClassificationService {
&& textBlock.getMostPopularWordStyle().equals("bold")
&& !document.getFontStyleCounter().getMostPopular().equals("bold")
&& PositionUtils.getApproxLineCount(textBlock) < 2.9
&& textBlock.getWords().get(0).getTextPositions().get(0).getFontSizeInPt() >= textBlock.getMostPopularWordFontSize()) {
&& textBlock.getWords().get(0).getFontSize() >= textBlock.getMostPopularWordFontSize()) {
PageBlockType headlineType = HeadlineClassificationService.headlineClassByFontSize(textBlock, headlineFontSizes);
headlineClassificationService.classifyHeadline(textBlock, headlineType);

12
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/factory/SearchTextWithTextPositionFactory.java
View File

@ -10,6 +10,7 @@ import java.util.List;
import java.util.Locale;
import java.util.Objects;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Character;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.TextRange;
import com.knecon.fforesight.service.layoutparser.processor.model.text.RedTextPosition;
import com.knecon.fforesight.service.layoutparser.processor.model.text.Word;
@ -31,19 +32,19 @@ public class SearchTextWithTextPositionFactory {
public SearchTextWithTextPositionDto buildSearchTextToTextPositionDto(List<Word> sequences) {
if (sequences.isEmpty() || sequences.stream()
.allMatch(sequence -> sequence.getTextPositions().isEmpty())) {
.allMatch(sequence -> sequence.getCharacters().isEmpty())) {
return SearchTextWithTextPositionDto.empty();
}
Context context = new Context();
RedTextPosition currentTextPosition = sequences.get(0).getTextPositions().get(0);
RedTextPosition currentTextPosition = sequences.get(0).getCharacters().get(0).getTextPosition();
RedTextPosition previousTextPosition = RedTextPosition.builder().unicode(" ").bBoxDirAdj(currentTextPosition.getBBoxDirAdj()).build();
for (Word word : sequences) {
for (int i = 0; i < word.getTextPositions().size(); ++i) {
for (int i = 0; i < word.getCharacters().size(); ++i) {
currentTextPosition = word.getTextPositions().get(i);
currentTextPosition = word.getCharacters().get(i).getTextPosition();
if (isLineBreak(currentTextPosition, previousTextPosition)) {
removeHyphenLinebreaks(context);
context.lineBreaksStringIdx.add(context.stringIdx);
@ -66,8 +67,9 @@ public class SearchTextWithTextPositionFactory {
}
List<Rectangle2D> positions = sequences.stream()
.map(Word::getTextPositions)
.map(Word::getCharacters)
.flatMap(Collection::stream)
.map(Character::getTextPosition)
.map(RedTextPosition::getBBoxPdf)
.toList();

4
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/parsing/PDFLinesTextStripper.java
View File

@ -248,8 +248,8 @@ public class PDFLinesTextStripper extends PDFTextStripper {
if (!words.isEmpty()) {
previous = words.get(words.size() - 1)
.getTextPositions()
.get(words.get(words.size() - 1).getTextPositions().size() - 1);
.getCharacters()
.get(words.get(words.size() - 1).getCharacters().size() - 1).getTextPosition();
}
if (i == 0 && (textPositions.get(i).getUnicode().equals(" ") || textPositions.get(i).getUnicode().equals("\u00A0") || textPositions.get(i).getUnicode().equals("\t"))) {

26
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/utils/TextPositionOperations.java
View File

@ -11,6 +11,7 @@ import java.util.Set;
import java.util.stream.Collectors;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.AngleFilter;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.Line;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.TextBoundingBox;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.model.UnionFind;
import com.knecon.fforesight.service.layoutparser.processor.docstrum.utils.DoubleUtils;
@ -54,11 +55,20 @@ public class TextPositionOperations {
private List<Word> sortUsingLineDetection(Set<Word> sequences) {
return sortLines(groupByLine(sequences));
return sortWords(groupByLine(sequences));
}
public List<Word> sortLines(Collection<Set<Word>> lines) {
public List<Word> sortWords(Collection<Set<Word>> lines) {
return sortLines(lines).stream()
.map(Line::getWords)
.flatMap(Collection::stream)
.collect(Collectors.toList());
}
public List<Line> sortLines(Collection<Set<Word>> lines) {
List<List<Word>> lineBlocks = new ArrayList<>();
for (Set<Word> line : lines) {
@ -70,9 +80,9 @@ public class TextPositionOperations {
// need to use old sorting, since COMPARATOR_DIR_ADJ is not transitive
QuickSort.sort(lineBlocks, Comparator.comparing(line -> line.get(0), COMPARATOR_DIR_ADJ));
List<Word> list = new ArrayList<>();
for (List<Word> words : lineBlocks) {
list.addAll(words);
List<Line> list = new ArrayList<>();
for (List<Word> lineBlock : lineBlocks) {
list.add(new Line(lineBlock));
}
return list;
}
@ -95,6 +105,12 @@ public class TextPositionOperations {
.map(Word::getBBoxDirAdj)
.mapToDouble(RectangularShape::getWidth).average().orElse(75) * MAX_WORD_DISTANCE_FACTOR;
return groupByLine(sequences, maxLineDistance, maxXGap);
}
public Collection<Set<Word>> groupByLine(Set<Word> sequences, double maxLineDistance, double maxXGap) {
UnionFind<Word> unionFind = new UnionFind<>(sequences);
for (Word sequence : sequences) {

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

@ -273,7 +273,9 @@ public class LayoutDebugLayer extends LayoutDebugLayerConfig {
AtomicInteger index = new AtomicInteger(0);
zones.forEach(zone -> zone.getLines()
.stream()
.map(Line::getCharacters)
.map(Line::getWords)
.flatMap(Collection::stream)
.map(Word::getCharacters)
.flatMap(Collection::stream)
.forEach(character -> {
Color color = getRotatingColor(index);