Merge branch 'RM-231-bp' into 'release/0.192.x'

RM-231: missing whitespace in name

See merge request fforesight/layout-parser!266

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

@@ -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);
 

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

@@ -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);

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

@@ -24,7 +24,7 @@ import lombok.EqualsAndHashCode;
 @EqualsAndHashCode(onlyExplicitlyIncluded = true, callSuper = false)
 public class Line extends TextBoundingBox {
 
-    private static final double WORD_DISTANCE_MULTIPLIER = 0.18;
+    private static final double WORD_DISTANCE_MULTIPLIER = 0.17;
 
     @EqualsAndHashCode.Include
     private final double x0;
@@ -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());
@@ -157,8 +155,11 @@ public class Line extends TextBoundingBox {
     }
 
 
-    private void computeWords(double wordSpacing) {
+    private void computeWords(List<Character> characters, double wordSpacing) {
 
+        // Imo, the width of space should be scaled with the font size, but it only depends on the median distance between horizontal neighbours.
+        // If there are large differences in fontsize on a page, this might lead to missing spaces for the smaller fonts and too many for larger fonts.
+        // I've just now changed the scaling factor. If you come across this comment with missing whitespaces again, try scaling the fontsize instead of simply changing the factor again.
         Word word = new Word();
         Character previous = null;
         for (Character current : characters) {
@@ -169,7 +170,7 @@ public class Line extends TextBoundingBox {
                     word = new Word();
                 }
             }
-            word.getTextPositions().add(current.getTextPosition());
+            word.add(current);
             previous = current;
         }
         words.add(word);
@@ -178,9 +179,7 @@ public class Line extends TextBoundingBox {
 
     private void buildBBox() {
 
-        this.setToBBoxOfComponents(characters.stream()
-                                           .map(Character::getTextPosition)
-                                           .toList());
+        this.setToBBoxOfComponents(words);
     }
 
 

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

@@ -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);
     }

layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/docstrum/service/LineBuilderService.java

@@ -17,7 +17,7 @@ public class LineBuilderService {
 
     private static final double CHARACTER_SPACING_DISTANCE_MULTIPLIER = 3.5;
     private static final double LINE_SPACING_THRESHOLD_MULTIPLIER = 0.67;
-    private static final double ANGLE_TOLERANCE = Math.PI / 6;
+    private static final double ANGLE_TOLERANCE = Math.toRadians(5);
 
 
     public List<Line> buildLines(List<Character> characters, double characterSpacing, double lineSpacing, CleanRulings rulings) {

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

@@ -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);
     }
 
 }

layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/model/SectionIdentifier.java

@@ -16,8 +16,8 @@ import lombok.experimental.FieldDefaults;
 @FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
 public class SectionIdentifier {
 
-    public static Pattern numericalIdentifierPattern = Pattern.compile("^[\\s]?(\\d+)[\\s.,;]?(\\d+)?[\\s.,;]?(\\d+)?[\\s.,;]?(\\d+)?");
-    public static Pattern alphanumericIdentifierPattern = Pattern.compile("^[\\s]?[A-Za-z][\\s.,;]?(\\d+)[\\s.,;]?(\\d+)?[\\s.,;]?(\\d+)?[\\s.,;]?(\\d+)?[\\s.,;]?");
+    public static Pattern numericalIdentifierPattern = Pattern.compile("^[\\s]?(\\d{1,2})(?:[\\s.,;](\\d{1,2}))?(?:[\\s.,;](\\d{1,2}))?(?:[\\s.,;](\\d{1,2}))?");
+    public static Pattern alphanumericIdentifierPattern = Pattern.compile("^[\\s]?[A-Za-z][\\s.,;]?(\\d{1,2})[\\s.,;]?(\\d{1,2})?[\\s.,;]?(\\d{1,2})?[\\s.,;]?(\\d{1,2})?[\\s.,;]?");
 
     public enum Format {
         EMPTY,

layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/model/text/ListIdentifier.java

@@ -14,11 +14,12 @@ import lombok.experimental.FieldDefaults;
 @FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
 public class ListIdentifier {
 
-    public static final Pattern NUMBER_WITH_DOT = Pattern.compile("^\\s*([1-9]+)\\.\\s+");
-    public static final Pattern NUMBER_IN_PARENTHESES = Pattern.compile("^\\s*\\(([1-9]+)\\)\\s+");
+    public static final Pattern NUMBER_WITH_DOT = Pattern.compile("^\\s*([1-9]{1,4})\\.\\s+");
+    public static final Pattern NUMBER_IN_PARENTHESES = Pattern.compile("^\\s*\\(([1-9]{1,4})\\)\\s+");
 
     enum Format {
-        NUMBER_WITH_DOT, NUMBER_IN_PARENTHESES
+        NUMBER_WITH_DOT,
+        NUMBER_IN_PARENTHESES
     }
 
     Format format;
@@ -48,17 +49,33 @@ public class ListIdentifier {
         Matcher numberMatcher = NUMBER_WITH_DOT.matcher(text);
 
         if (numberMatcher.find()) {
-            return Optional.of(new ListIdentifier(Format.NUMBER_WITH_DOT, sequences.get(0), page, Integer.parseInt(numberMatcher.group(1))));
+            Optional<Integer> representation = parseInteger(numberMatcher.group(1));
+            if (representation.isPresent()) {
+                return Optional.of(new ListIdentifier(Format.NUMBER_WITH_DOT, sequences.get(0), page, representation.get()));
+            }
         }
 
         Matcher parenthesisMatcher = NUMBER_IN_PARENTHESES.matcher(text);
         if (parenthesisMatcher.find()) {
-            return Optional.of(new ListIdentifier(Format.NUMBER_IN_PARENTHESES, sequences.get(0), page, Integer.parseInt(parenthesisMatcher.group(1))));
+            Optional<Integer> representation = parseInteger(parenthesisMatcher.group(1));
+            if (representation.isPresent()) {
+                return Optional.of(new ListIdentifier(Format.NUMBER_IN_PARENTHESES, sequences.get(0), page, representation.get()));
+            }
         }
         return Optional.empty();
     }
 
 
+    private static Optional<Integer> parseInteger(String text) {
+
+        try {
+            return Optional.of(Integer.parseInt(text));
+        } catch (NumberFormatException e) {
+            return Optional.empty();
+        }
+    }
+
+
     public static boolean isInOrder(List<ListIdentifier> listIdentifiers) {
 
         if (listIdentifiers.size() <= 1) {

layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/model/text/TextPositionSequenceComparator.java

@@ -1,34 +0,0 @@
-package com.knecon.fforesight.service.layoutparser.processor.model.text;
-
-import java.util.Comparator;
-import java.util.HashMap;
-
-public class TextPositionSequenceComparator implements Comparator<Word> {
-
-    private HashMap<Word, TextBlockOnPage> lookup;
-
-
-    public TextPositionSequenceComparator(HashMap<Word, TextBlockOnPage> lookup) {
-
-        this.lookup = lookup;
-    }
-
-
-    @Override
-    public int compare(Word number1, Word number2) {
-
-        int page1 = lookup.get(number1).page().getPageNumber();
-        int page2 = lookup.get(number2).page().getPageNumber();
-
-        if (page1 != page2) {
-            return Integer.compare(page1, page2);
-        }
-
-        if (number1.getY() != number2.getY()) {
-            return Double.compare(number1.getY(), number2.getY());
-        }
-
-        return Integer.compare(Integer.parseInt(number1.toString()), Integer.parseInt(number2.toString()));
-    }
-
-}

layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/model/text/TocNumberComparator.java

@@ -0,0 +1,36 @@
+package com.knecon.fforesight.service.layoutparser.processor.model.text;
+
+import java.util.Comparator;
+import java.util.HashMap;
+
+import com.knecon.fforesight.service.layoutparser.processor.services.classification.NumberWord;
+
+public class TocNumberComparator implements Comparator<NumberWord> {
+
+    private HashMap<NumberWord, TextBlockOnPage> lookup;
+
+
+    public TocNumberComparator(HashMap<NumberWord, TextBlockOnPage> lookup) {
+
+        this.lookup = lookup;
+    }
+
+
+    @Override
+    public int compare(NumberWord number1, NumberWord number2) {
+
+        int page1 = lookup.get(number1).page().getPageNumber();
+        int page2 = lookup.get(number2).page().getPageNumber();
+
+        if (page1 != page2) {
+            return Integer.compare(page1, page2);
+        }
+
+        if (number1.word().getY() != number2.word().getY()) {
+            return Double.compare(number1.word().getY(), number2.word().getY());
+        }
+
+        return Integer.compare(number1.number(), number2.number());
+    }
+
+}

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

@@ -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()) {

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

@@ -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;
     }

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

@@ -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()));
                                 });
                     });
 

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

@@ -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);

layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/classification/NumberWord.java

@@ -0,0 +1,7 @@
+package com.knecon.fforesight.service.layoutparser.processor.services.classification;
+
+import com.knecon.fforesight.service.layoutparser.processor.model.text.Word;
+
+public record NumberWord(Word word, int number) {
+
+}

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

@@ -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);

layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/classification/TableOfContentsClassificationService.java

@@ -14,6 +14,7 @@ import java.util.List;
 import java.util.Locale;
 import java.util.Map;
 import java.util.Set;
+import java.util.regex.Matcher;
 import java.util.stream.Collectors;
 
 import org.springframework.stereotype.Service;
@@ -23,10 +24,11 @@ import com.knecon.fforesight.service.layoutparser.processor.model.AbstractPageBl
 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.text.TextBlockOnPage;
 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.model.text.TextPositionSequenceComparator;
+import com.knecon.fforesight.service.layoutparser.processor.model.text.TocNumberComparator;
 import com.knecon.fforesight.service.layoutparser.processor.utils.TextNormalizationUtilities;
 import com.knecon.fforesight.service.layoutparser.processor.visualization.LayoutDebugLayer;
 
@@ -69,11 +71,14 @@ public class TableOfContentsClassificationService {
 
     private int identifyTOCItems(int start, List<TextBlockOnPage> textBlocks, ClassificationDocument document) {
 
+        if (start >= textBlocks.size()) {
+            return start;
+        }
         ClassificationPage startPage = textBlocks.get(start).page();
         List<TextBlockOnPage> initialLookAhead = textBlocks.subList(start, Math.min(start + SURROUNDING_BLOCKS_RADIUS, textBlocks.size()));
-        HashMap<Word, TextBlockOnPage> lookup = new HashMap<>();
-        List<Word> numbers = extractNumbers(initialLookAhead, lookup, document.getPages().size());
-        TocNumberFinder tocNumberFinder = new TocNumberFinder(numbers, lookup);
+        HashMap<NumberWord, TextBlockOnPage> numberToBlockLookup = new HashMap<>();
+        List<NumberWord> numbers = extractNumbers(initialLookAhead, numberToBlockLookup, document.getPages().size());
+        TocNumberFinder tocNumberFinder = new TocNumberFinder(numbers, numberToBlockLookup);
 
         int lastCandidate = start;
         for (int i = start; i < Math.min(lastCandidate + SURROUNDING_BLOCKS_RADIUS, textBlocks.size()); i++) {
@@ -93,28 +98,28 @@ public class TableOfContentsClassificationService {
                 break;
             }
 
-            List<Word> numbersFromBlock = extractNumbers(textBlockOnPage, lookup, document.getPages().size());
+            List<NumberWord> numbersFromBlock = extractNumbers(textBlockOnPage, numberToBlockLookup, document.getPages().size());
 
-            List<Word> currentRightmostCluster = tocNumberFinder.getCurrentRightmostCluster();
+            List<NumberWord> currentRightmostCluster = tocNumberFinder.getCurrentRightmostCluster();
 
             if (currentRightmostCluster.size() < MINIMUM_MATCHES) {
                 log.debug("No numbers indicating a table of contents here.");
                 return start;
             }
 
-            if (anyIntersection(currentRightmostCluster, numbersFromBlock, lookup)) {
+            if (anyIntersection(currentRightmostCluster, numbersFromBlock, numberToBlockLookup)) {
                 lastCandidate = i;
                 numbersFromBlock.forEach(tocNumberFinder::add);
             }
         }
 
-        addVisualization(document.getLayoutDebugLayer(), tocNumberFinder, lookup);
-
         Set<TextBlockOnPage> blocksWithNumberInCluster = tocNumberFinder.getCurrentRightmostCluster()
                 .stream()
-                .map(lookup::get)
+                .map(numberToBlockLookup::get)
                 .collect(Collectors.toSet());
 
+        addVisualization(document.getLayoutDebugLayer(), tocNumberFinder, numberToBlockLookup, blocksWithNumberInCluster, textBlocks.get(start - 1));
+
         int lastConfirmed = start;
         for (int i = start; i < lastCandidate + 1; i++) {
             TextBlockOnPage textBlockOnPage = textBlocks.get(i);
@@ -132,18 +137,22 @@ public class TableOfContentsClassificationService {
     }
 
 
-    private static void addVisualization(LayoutDebugLayer layoutDebugLayer, TocNumberFinder tocNumberFinder, Map<Word, TextBlockOnPage> lookup) {
+    private static void addVisualization(LayoutDebugLayer layoutDebugLayer,
+                                         TocNumberFinder tocNumberFinder,
+                                         Map<NumberWord, TextBlockOnPage> lookup,
+                                         Set<TextBlockOnPage> blocksWithNumberInCluster,
+                                         TextBlockOnPage startingHeadline) {
 
         tocNumberFinder.getCurrentRightmostCluster()
                 .stream()
                 .collect(Collectors.groupingBy(key -> lookup.get(key).page().getPageNumber()))
                 .forEach((pageNumber, number) -> layoutDebugLayer.addTocPages(number, pageNumber));
+        layoutDebugLayer.addTocBlocks(blocksWithNumberInCluster);
+        layoutDebugLayer.addTocBlocks(Set.of(startingHeadline));
     }
 
 
-    private static boolean anyIntersection(Collection<Word> numbers1,
-                                           Collection<Word> numbers2,
-                                           Map<Word, TextBlockOnPage> lookup) {
+    private static boolean anyIntersection(Collection<NumberWord> numbers1, Collection<NumberWord> numbers2, Map<NumberWord, TextBlockOnPage> lookup) {
 
         return numbers1.stream()
                 .anyMatch(numberFromCluster -> numbers2.stream()
@@ -151,9 +160,9 @@ public class TableOfContentsClassificationService {
     }
 
 
-    private static List<Word> extractNumbers(List<TextBlockOnPage> textBlocks, Map<Word, TextBlockOnPage> lookup, int numberOfPages) {
+    private static List<NumberWord> extractNumbers(List<TextBlockOnPage> textBlocks, Map<NumberWord, TextBlockOnPage> lookup, int numberOfPages) {
 
-        List<Word> blocks = new LinkedList<>();
+        List<NumberWord> blocks = new LinkedList<>();
         for (TextBlockOnPage textBlock : textBlocks) {
             blocks.addAll(extractNumbers(textBlock, lookup, numberOfPages));
         }
@@ -161,30 +170,40 @@ public class TableOfContentsClassificationService {
     }
 
 
-    private static List<Word> extractNumbers(TextBlockOnPage textBlock, Map<Word, TextBlockOnPage> lookup, int numberOfPages) {
+    private static List<NumberWord> extractNumbers(TextBlockOnPage textBlock, Map<NumberWord, TextBlockOnPage> lookup, int numberOfPages) {
 
-        List<Word> blocks = new LinkedList<>();
+        List<NumberWord> blocks = new LinkedList<>();
         TextPageBlock block = textBlock.textBlock();
-        List<Word> sequences = block.getWords();
-        for (int i = 0; i < sequences.size(); i++) {
+        List<Word> words = block.getWords();
+        for (int i = 0; i < words.size(); i++) {
 
-            Word word = sequences.get(i);
-
-            if (!NUMERIC.matcher(word).matches()) {
+            Word word = words.get(i);
+            if (!wordIsEndOfLine(i, words)) {
                 continue;
             }
 
-            if (AMOUNT_PATTERN.matcher(getSurroundingString(i, sequences)).matches()) {
+            if (AMOUNT_PATTERN.matcher(getSurroundingString(i, words)).matches()) {
+                continue;
+            }
+
+            Matcher matcher = SectionIdentifier.numericalIdentifierPattern.matcher(word.toString());
+            if (matcher.find() && matcher.group(2) != null) {
+                continue;
+            }
+
+            Matcher numberFinder = NUMERIC.matcher(word);
+            if (!numberFinder.find() || word.length() > 5) {
                 continue;
             }
 
             try {
-                int pageNumber = Integer.parseInt(word.toString());
+                int pageNumber = Integer.parseInt(numberFinder.group());
                 if (0 >= pageNumber || pageNumber > numberOfPages) {
                     continue;
                 }
-                lookup.put(word, textBlock);
-                blocks.add(word);
+                NumberWord numberWord = new NumberWord(word, pageNumber);
+                lookup.put(numberWord, textBlock);
+                blocks.add(numberWord);
             } catch (NumberFormatException e) {
                 log.debug("That wasn't a number! Should not happen, due to numeric check beforehand.");
             }
@@ -193,6 +212,17 @@ public class TableOfContentsClassificationService {
     }
 
 
+    private static boolean wordIsEndOfLine(int i, List<Word> words) {
+
+        if (i == words.size() - 1) {
+            return true;
+        }
+        Word word = words.get(i);
+        Word nextWord = words.get(i + 1);
+        return !nextWord.rightOf(word);
+    }
+
+
     private static CharSequence getSurroundingString(int i, List<Word> sequences) {
 
         int end = Math.min(i + 5, sequences.size());
@@ -203,13 +233,13 @@ public class TableOfContentsClassificationService {
     }
 
 
-    private static boolean matches(Word number1, Word number2, Map<Word, TextBlockOnPage> lookup) {
+    private static boolean matches(NumberWord number1, NumberWord number2, Map<NumberWord, TextBlockOnPage> lookup) {
 
-        if (number1.getDir() != number2.getDir()) {
+        if (number1.word().getDir() != number2.word().getDir()) {
             return false;
         }
 
-        return number1.intersectsXDirAdj(number2, INTERSECTION_TOLERANCE);
+        return number1.word().intersectsXDirAdj(number2.word(), INTERSECTION_TOLERANCE);
     }
 
 
@@ -247,11 +277,11 @@ public class TableOfContentsClassificationService {
 
     private static class TocNumberFinder {
 
-        final UnionFind<Word> numberClusters;
-        final HashMap<Word, TextBlockOnPage> lookup;
+        final UnionFind<NumberWord> numberClusters;
+        final HashMap<NumberWord, TextBlockOnPage> lookup;
 
 
-        TocNumberFinder(List<Word> blocks, HashMap<Word, TextBlockOnPage> lookup) {
+        TocNumberFinder(List<NumberWord> blocks, HashMap<NumberWord, TextBlockOnPage> lookup) {
 
             this.numberClusters = new UnionFind<>(new HashSet<>(blocks));
             for (int i = 0; i < blocks.size(); i++) {
@@ -265,14 +295,14 @@ public class TableOfContentsClassificationService {
         }
 
 
-        public void add(Word number) {
+        public void add(NumberWord number) {
 
             if (numberClusters.getElements().contains(number)) {
                 return;
             }
 
             numberClusters.addElement(number);
-            for (Word element : numberClusters.getElements()) {
+            for (NumberWord element : numberClusters.getElements()) {
                 if (matches(number, element, lookup)) {
                     numberClusters.union(element, number);
                 }
@@ -280,73 +310,100 @@ public class TableOfContentsClassificationService {
         }
 
 
-        public List<Word> getCurrentRightmostCluster() {
+        public List<NumberWord> getCurrentRightmostCluster() {
 
             return numberClusters.getGroups()
                     .stream()
                     .filter(cluster -> cluster.size() > MINIMUM_MATCHES)
                     .map(cluster -> cluster.stream()
-                            .sorted(new TextPositionSequenceComparator(lookup))
+                            .sorted(new TocNumberComparator(lookup))
                             .toList())
                     .map(this::removeOutliers)
-//                    .map(this::filterByMinimumDensity)
+                    .map(this::removeOnNonConsecutivePages)
+                    .map(this::filterByWordNearTopOfPage)
                     .filter(cluster -> cluster.size() > MINIMUM_MATCHES)
-                    .max(Comparator.comparingDouble(cluster -> cluster.get(0).getBBox().getMaxX())).orElse(Collections.emptyList());
+                    .max(Comparator.comparingDouble(cluster -> cluster.get(0).word().getBBox().getMaxX())).orElse(Collections.emptyList());
         }
 
-// does not seem to be doing much, ideally instead of using the height of the blocks i would like to use the height, beginning from the MainBody top,
-// but as the MainBody is often wrong, this results in some numbers being discarded even though they are correct.
-//        private List<TextPositionSequence> filterByMinimumDensity(List<TextPositionSequence> numbers) {
-//
-//            Map<ClassificationPage, List<TextPositionSequence>> clustersPerPage = numbers.stream()
-//                    .collect(Collectors.groupingBy(number -> lookup.get(number).page()));
-//
-//            List<TextPositionSequence> result = new ArrayList<>(numbers.size());
-//            clustersPerPage.keySet()
-//                    .stream()
-//                    .sorted(Comparator.comparingInt(ClassificationPage::getPageNumber))
-//                    .forEach(page -> {
-//                        var numbersOnPage = clustersPerPage.get(page);
-//
-//                        double height = numbersOnPage.stream()
-//                                .map(BoundingBox::getBBox)
-//                                .collect(RectangleTransformations.collectBBox()).getHeight();
-//
-//                        double count = numbersOnPage.size();
-//
-//                        if ((count / height) >= (DENSITY_THRESHOLD_COUNT / page.getPageHeight())) {
-//                            result.addAll(numbers);
-//                        }
-//                    });
-//            return result;
-//        }
 
+        private List<NumberWord> removeOnNonConsecutivePages(List<NumberWord> numbers) {
 
-        public List<Word> removeOutliers(List<Word> numbers) {
-
-            List<Word> result = new ArrayList<>();
+            List<NumberWord> result = new ArrayList<>();
 
             result.add(numbers.get(0));
 
+            for (int i = 1; i < numbers.size(); i++) {
+                int prev = getPageNumber(numbers, i - 1);
+                int curr = getPageNumber(numbers, i);
+
+                if (Math.abs(prev - curr) > 1) {
+                    break;
+                } else {
+                    result.add(numbers.get(i));
+                }
+            }
+            return result;
+        }
+
+
+        private int getPageNumber(List<NumberWord> numbers, int i) {
+
+            return lookup.get(numbers.get(i)).page().getPageNumber();
+        }
+
+
+        private List<NumberWord> filterByWordNearTopOfPage(List<NumberWord> numbers) {
+
+            List<NumberWord> result = new ArrayList<>();
+
+            result.add(numbers.get(0));
+
+            for (int i = 1; i < numbers.size(); i++) {
+                NumberWord prev = numbers.get(i - 1);
+                NumberWord curr = numbers.get(i);
+                ClassificationPage prevPage = lookup.get(prev).page();
+                ClassificationPage currPage = lookup.get(curr).page();
+                if (prevPage.equals(currPage)) {
+                    result.add(curr);
+                } else if (curr.word().getBBox().getMinY() < currPage.getPageHeight() * 0.33) {
+                    result.add(curr);
+                }
+            }
+            return result;
+        }
+
+
+        public List<NumberWord> removeOutliers(List<NumberWord> numbers) {
+
+            List<NumberWord> confirmedClusterNumbers = new ArrayList<>();
+
+            confirmedClusterNumbers.add(numbers.get(0));
+
             for (int i = 1; i < numbers.size() - 1; i++) {
                 int prev = getNumberAsInt(numbers, i - 1);
                 int curr = getNumberAsInt(numbers, i);
                 int next = getNumberAsInt(numbers, i + 1);
 
                 if (!(curr <= prev || curr >= next) || !isBetterWithout(numbers, i)) {
-                    result.add(numbers.get(i));
+                    confirmedClusterNumbers.add(numbers.get(i));
                 }
             }
-            if (getNumberAsInt(numbers, numbers.size() - 1) >= getNumberAsInt(numbers, Math.max(0, numbers.size() - 2))) {
-                result.add(numbers.get(numbers.size() - 1));
+            if (getNumberAsInt(numbers, numbers.size() - 1) >= getLatestNumber(confirmedClusterNumbers)) {
+                confirmedClusterNumbers.add(numbers.get(numbers.size() - 1));
             }
 
-            return result;
+            return confirmedClusterNumbers;
+        }
+
+
+        private static int getLatestNumber(List<NumberWord> confirmedClusterNumbers) {
+
+            return confirmedClusterNumbers.get(confirmedClusterNumbers.size() - 1).number();
         }
 
 
         // Helper method to check if removing the current number results in a better order
-        public static boolean isBetterWithout(List<Word> numbers, int i) {
+        public static boolean isBetterWithout(List<NumberWord> numbers, int i) {
 
             if (i == 0 || i == numbers.size() - 1) {
                 return false;
@@ -362,9 +419,9 @@ public class TableOfContentsClassificationService {
     }
 
 
-    private static int getNumberAsInt(List<Word> numbers, int i) {
+    private static int getNumberAsInt(List<NumberWord> numbers, int i) {
 
-        return Integer.parseInt(numbers.get(i).toString());
+        return numbers.get(i).number();
     }
 
 }

layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/factory/DocumentGraphFactory.java

@@ -16,7 +16,7 @@ import java.util.Optional;
 import java.util.Set;
 import java.util.stream.Collectors;
 
-import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeType;
+import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.NodeTypeProto;
 import com.knecon.fforesight.service.layoutparser.internal.api.queue.LayoutParsingType;
 import com.knecon.fforesight.service.layoutparser.processor.model.AbstractPageBlock;
 import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationDocument;
@@ -82,9 +82,9 @@ public class DocumentGraphFactory {
 
         documentGraph.streamAllSubNodes()
                 .filter(SemanticNode::isLeaf)
-                .filter(node -> !node.getType().equals(NodeType.HEADER))
-                .filter(node -> !node.getType().equals(NodeType.FOOTER))
-                .filter(node -> !node.getType().equals(NodeType.IMAGE))
+                .filter(node -> !node.getType().equals(NodeTypeProto.NodeType.HEADER))
+                .filter(node -> !node.getType().equals(NodeTypeProto.NodeType.FOOTER))
+                .filter(node -> !node.getType().equals(NodeTypeProto.NodeType.IMAGE))
                 .map(SemanticNode::getTextBlock)
                 .map(TextBlock::getAtomicTextBlocks)
                 .flatMap(Collection::stream)

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

@@ -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();
 

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

@@ -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"))) {

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

@@ -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) {

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

@@ -27,7 +27,9 @@ import com.knecon.fforesight.service.layoutparser.processor.model.outline.Outlin
 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.ListIdentifier;
+import com.knecon.fforesight.service.layoutparser.processor.model.text.TextBlockOnPage;
 import com.knecon.fforesight.service.layoutparser.processor.model.text.Word;
+import com.knecon.fforesight.service.layoutparser.processor.services.classification.NumberWord;
 import com.knecon.fforesight.service.layoutparser.processor.utils.CoordinateTransforms;
 import com.knecon.fforesight.service.layoutparser.processor.utils.MarkedContentUtils;
 import com.knecon.fforesight.service.layoutparser.processor.utils.PageInformation;
@@ -273,7 +275,9 @@ 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);
@@ -291,7 +295,7 @@ public class LayoutDebugLayer extends LayoutDebugLayerConfig {
     }
 
 
-    public void addTocPages(List<Word> numbers, int page) {
+    public void addTocPages(List<NumberWord> numbers, int page) {
 
         if (!active) {
             return;
@@ -300,13 +304,10 @@ public class LayoutDebugLayer extends LayoutDebugLayerConfig {
         VisualizationsOnPage visualizationsOnPage = getOrCreateVisualizationsOnPage(page, this.tocPages);
         visualizationsOnPage.getColoredRectangles()
                 .addAll(numbers.stream()
+                                .map(NumberWord::word)
                                 .map(BoundingBox::getBBoxPdf)
                                 .map(line -> new ColoredRectangle(line, LINES_COLOR, LINE_WIDTH))
                                 .toList());
-        visualizationsOnPage.getColoredRectangles()
-                .add(new ColoredRectangle(numbers.stream()
-                                                  .map(BoundingBox::getBBoxPdf)
-                                                  .collect(RectangleTransformations.collectBBox()), LINES_COLOR, LINE_WIDTH));
     }
 
 
@@ -330,8 +331,10 @@ public class LayoutDebugLayer extends LayoutDebugLayerConfig {
 
     private void addOutlineObject(OutlineObject outlineObject, PageInformation pageInformation) {
 
+        if (!active) {
+            return;
+        }
         int rectSize = 5;
-
         Point2D point2D;
         if (outlineObject.getPoint().isPresent()) {
             point2D = outlineObject.getPoint().get();
@@ -355,10 +358,25 @@ public class LayoutDebugLayer extends LayoutDebugLayerConfig {
 
     public void addListIdentifiers(List<ListIdentifier> listIdentifiers) {
 
+        if (!active) {
+            return;
+        }
         for (ListIdentifier listIdentifier : listIdentifiers) {
             getOrCreateVisualizationsOnPage(listIdentifier.getPage(), this.listIdentifiers).getColoredRectangles()
                     .add(new ColoredRectangle(listIdentifier.getWord().getBBoxPdf(), WORDS_COLOR, LINE_WIDTH));
         }
     }
 
+
+    public void addTocBlocks(Set<TextBlockOnPage> blocksWithNumberInCluster) {
+
+        if (!active) {
+            return;
+        }
+        for (TextBlockOnPage textBlockOnPage : blocksWithNumberInCluster) {
+            getOrCreateVisualizationsOnPage(textBlockOnPage.page().getPageNumber(), this.tocBlocks).getColoredRectangles()
+                    .add(new ColoredRectangle(textBlockOnPage.textBlock().getBBoxPdf(), TOC_COLOR, LINE_WIDTH));
+        }
+    }
+
 }

layoutparser-service/layoutparser-service-server/src/test/java/com/knecon/fforesight/service/layoutparser/processor/model/SectionIdentifierTest.java

@@ -83,4 +83,11 @@ class SectionIdentifierTest {
         assertEquals(List.of(4, 1, 2, 4), identifier.getIdentifiers());
     }
 
+    @Test
+    void testFalsePositive111() {
+        SectionIdentifier identifier = SectionIdentifier.fromSearchText("111: Headline");
+        assertEquals(SectionIdentifier.Format.NUMERICAL, identifier.getFormat());
+        assertEquals(1, identifier.level());
+    }
+
 }

layoutparser-service/viewer-doc-processor/src/main/java/com/knecon/fforesight/service/viewerdoc/LayerIdentifier.java

@@ -57,6 +57,7 @@ public record LayerIdentifier(String name, String markedContentName) {
     public static final LayerIdentifier OUTLINE_OBJECTS = new LayerIdentifier("Outline Positions", "OUTLINE_OBJECTS");
     public static final LayerIdentifier SENTENCES = new LayerIdentifier("Sentences", "SENTENCES");
     public static final LayerIdentifier TOC_PAGES = new LayerIdentifier("TOC pages", "TOC_PAGES");
+    public static final LayerIdentifier TOC_BLOCKS = new LayerIdentifier("TOC blocks", "TOC_BLOCKS");
     public static final LayerIdentifier LIST_IDENTIFIERS = new LayerIdentifier("List identifiers", "LIST_IDENTIFIERS");
 
     // Visual layout parser

layoutparser-service/viewer-doc-processor/src/main/java/com/knecon/fforesight/service/viewerdoc/layers/LayoutDebugLayerConfig.java

@@ -22,6 +22,7 @@ public class LayoutDebugLayerConfig extends AbstractLayerGroup {
 
     protected static final Color WORDS_COLOR = new Color(68, 84, 147);
     protected static final Color LINES_COLOR = new Color(152, 45, 179);
+    protected static final Color TOC_COLOR = new Color(33, 159, 144);
     protected static final Color ZONES_COLOR = new Color(131, 38, 38);
 
     protected static final Color RULINGS_COLOR = new Color(21, 221, 174);
@@ -59,6 +60,7 @@ public class LayoutDebugLayerConfig extends AbstractLayerGroup {
     protected final Visualizations outlineObjects = Visualizations.builder().layer(LayerIdentifier.OUTLINE_OBJECTS).build();
     protected final Visualizations sentences = Visualizations.builder().layer(LayerIdentifier.SENTENCES).build();
     protected final Visualizations tocPages = Visualizations.builder().layer(LayerIdentifier.TOC_PAGES).build();
+    protected final Visualizations tocBlocks = Visualizations.builder().layer(LayerIdentifier.TOC_BLOCKS).build();
     protected final Visualizations listIdentifiers = Visualizations.builder().layer(LayerIdentifier.LIST_IDENTIFIERS).build();
 
 
@@ -77,9 +79,9 @@ public class LayoutDebugLayerConfig extends AbstractLayerGroup {
                        markedContent, //
                        outlineObjects, //
                        tocPages, //
+                       tocBlocks, //
                        listIdentifiers //
         );
     }
 
-
 }