RED-9964: don't merge tables on non-consecutive pages

This commit is contained in:
Kilian Schüttler 2024-08-30 14:00:48 +02:00
parent cf39d4dfcc
commit c5178ea5c2
10 changed files with 182 additions and 103 deletions

View File

@ -133,7 +133,7 @@ public abstract class BoundingBox {
}
private boolean intersectsX(BoundingBox other, float threshold) {
public boolean intersectsX(BoundingBox other, float threshold) {
return this.getX() - threshold <= other.getMaxX() && this.getMaxX() + threshold >= other.getX();
}

View File

@ -1,12 +1,15 @@
package com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes;
import java.util.Comparator;
import java.util.HashSet;
import java.util.LinkedList;
import java.util.List;
import java.util.Set;
import java.util.stream.Stream;
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationPage;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.entity.TextEntity;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.AtomicTextBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.TextBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.TextBlockCollector;
@ -29,9 +32,8 @@ public class Page {
Integer height;
Integer width;
Integer rotation;
@EqualsAndHashCode.Exclude
List<SemanticNode> mainBody;
List<AtomicTextBlock> textBlocksOnPage;
@EqualsAndHashCode.Exclude
Header header;
@EqualsAndHashCode.Exclude
@ -53,20 +55,43 @@ public class Page {
.width((int) classificationPage.getPageWidth())
.number(classificationPage.getPageNumber())
.rotation(classificationPage.getRotation())
.mainBody(new LinkedList<>())
.textBlocksOnPage(new LinkedList<>())
.build();
}
/**
* Constructs and returns a {@link TextBlock} representing the concatenated text of all leaf semantic nodes in the main body.
*
* @return The main body text block.
*/
public TextBlock getMainBodyTextBlock() {
return mainBody.stream()
.filter(SemanticNode::isLeaf)
.map(SemanticNode::getLeafTextBlock)
return textBlocksOnPage.stream()
.collect(new TextBlockCollector());
}
public List<SemanticNode> getMainBody() {
return textBlocksOnPage.stream()
.map(AtomicTextBlock::getParent)
.map(this::getHighestParentOnPage)
.distinct()
.toList();
}
private SemanticNode getHighestParentOnPage(SemanticNode node) {
SemanticNode currentNode = node;
while (currentNode.getParent().onlyOnPage(this)) {
currentNode = currentNode.getParent();
}
return currentNode;
}
@Override
public String toString() {

View File

@ -74,7 +74,8 @@ public interface SemanticNode {
return getTextBlock().getPages()
.stream()
.min(Comparator.comparingInt(Page::getNumber)).orElseThrow(() -> new IllegalStateException("SemanticNode has no Page!"));
.min(Comparator.comparingInt(Page::getNumber))
.orElseThrow(() -> new IllegalStateException("SemanticNode has no Page!"));
}
@ -504,4 +505,17 @@ public interface SemanticNode {
void accept(NodeVisitor visitor);
/**
* Checks wether this SemanticNode appears on a single page only, and if that page is the provided one.
*
* @param page the page to check
* @return true, when SemanticNode is on a single page only and the page is the provided page. Otherwise, false.
*/
default boolean onlyOnPage(Page page) {
Set<Page> pages = getPages();
return pages.size() == 1 && pages.contains(page);
}
}

View File

@ -6,6 +6,7 @@ import static java.util.stream.Collectors.toList;
import java.awt.geom.Rectangle2D;
import java.util.ArrayList;
import java.util.Collection;
import java.util.HashMap;
import java.util.LinkedList;
import java.util.List;
@ -15,6 +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.queue.LayoutParsingType;
import com.knecon.fforesight.service.layoutparser.processor.model.AbstractPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationDocument;
@ -32,7 +34,9 @@ import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.He
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Image;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Page;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.Paragraph;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.nodes.SemanticNode;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.AtomicTextBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.graph.textblock.TextBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.image.ClassifiedImage;
import com.knecon.fforesight.service.layoutparser.processor.model.outline.TableOfContentItem;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageBlock;
@ -68,10 +72,25 @@ public class DocumentGraphFactory {
documentGraph.setPages(context.pages.keySet());
documentGraph.setDocumentTree(context.documentTree);
documentGraph.setTextBlock(documentGraph.getTextBlock());
addTextBlocksToPages(documentGraph);
return documentGraph;
}
private void addTextBlocksToPages(Document documentGraph) {
documentGraph.streamAllSubNodes()
.filter(SemanticNode::isLeaf)
.filter(node -> !node.getType().equals(NodeType.HEADER))
.filter(node -> !node.getType().equals(NodeType.FOOTER))
.map(SemanticNode::getTextBlock)
.map(TextBlock::getAtomicTextBlocks)
.flatMap(Collection::stream)
.forEach(atb -> atb.getPage().getTextBlocksOnPage().add(atb));
}
private void addSections(LayoutParsingType layoutParsingType, ClassificationDocument classificationDocument, Context context, Document document) {
for (TableOfContentItem tocItem : classificationDocument.getTableOfContents()) {
@ -105,8 +124,6 @@ public class DocumentGraphFactory {
node = Paragraph.builder().documentTree(context.getDocumentTree()).build();
}
page.getMainBody().add(node);
List<TextPageBlock> textBlocks = new ArrayList<>();
textBlocks.add(originalTextBlock);
textBlocks.addAll(textBlocksToMerge);
@ -115,9 +132,9 @@ public class DocumentGraphFactory {
if (node instanceof DuplicatedParagraph duplicatedParagraph) {
AtomicTextBlock unsortedTextBlock = context.textBlockFactory.buildAtomicTextBlock2(textBlocks.stream()
.flatMap(tb -> tb.getSequences()
.stream())
.collect(Collectors.toList()), node, context, page);
.flatMap(tb -> tb.getSequences()
.stream())
.collect(Collectors.toList()), node, context, page);
duplicatedParagraph.setUnsortedLeafTextBlock(unsortedTextBlock);
}
@ -141,7 +158,7 @@ public class DocumentGraphFactory {
Rectangle2D position = image.getPosition();
Page page = context.getPage(image.getPage());
Image imageNode = Image.builder()
return Image.builder()
.id(IdBuilder.buildId(Set.of(page), List.of(position)))
.imageType(image.getImageType())
.position(position)
@ -150,8 +167,6 @@ public class DocumentGraphFactory {
.representationHash(image.getRepresentation())
.documentTree(context.getDocumentTree())
.build();
page.getMainBody().add(imageNode);
return imageNode;
}

View File

@ -51,9 +51,6 @@ public class SectionNodeFactory {
return Optional.empty();
}
Map<Integer, List<AbstractPageBlock>> blocksPerPage = pageBlocks.stream()
.collect(groupingBy(AbstractPageBlock::getPage));
AbstractSemanticNode section;
boolean containsTablesAndTextBlocks = containsTablesAndTextBlocks(pageBlocks);
if (isLeaf && !containsTablesAndTextBlocks) {
@ -63,8 +60,6 @@ public class SectionNodeFactory {
}
context.getSections().add(section);
blocksPerPage.keySet()
.forEach(pageNumber -> addSectionNodeToPageNode(context, section, pageNumber));
section.setTreeId(getTreeId(parentNode, context, section));
@ -242,10 +237,5 @@ public class SectionNodeFactory {
}
private void addSectionNodeToPageNode(DocumentGraphFactory.Context context, AbstractSemanticNode section, Integer pageNumber) {
Page page = context.getPage(pageNumber);
page.getMainBody().add(section);
}
}

View File

@ -51,8 +51,6 @@ public class TableNodeFactory {
.numberOfRows(mergedRows.size())
.build();
pages.forEach(page -> addTableToPage(page, parentNode, table));
List<Integer> treeId = context.getDocumentTree().createNewChildEntryAndReturnId(parentNode, table);
table.setTreeId(treeId);
addTableCells(layoutParsingType, mergedRows, table, context, document);
@ -82,17 +80,6 @@ public class TableNodeFactory {
}
@SuppressWarnings("PMD.UnusedPrivateMethod") // PMD actually flags this wrong
private void addTableToPage(Page page, SemanticNode parentNode, Table table) {
if (!page.getMainBody().contains(parentNode)) {
parentNode.getPages().add(page);
}
page.getMainBody().add(table);
}
private void ifTableHasNoHeadersSetFirstRowAsHeaders(Table table) {
if (table.streamHeaders()
@ -107,14 +94,7 @@ public class TableNodeFactory {
for (int rowIndex = 0; rowIndex < rows.size(); rowIndex++) {
for (int colIndex = 0; colIndex < rows.get(rowIndex).size(); colIndex++) {
addTableCell(layoutParsingType,
rows.get(rowIndex)
.get(colIndex),
rowIndex,
colIndex,
table,
context,
document);
addTableCell(layoutParsingType, rows.get(rowIndex).get(colIndex), rowIndex, colIndex, table, context, document);
}
}
}
@ -131,14 +111,7 @@ public class TableNodeFactory {
Page page = context.getPage(cell.getPageNumber());
TableCell tableCell = TableCell.builder()
.documentTree(context.getDocumentTree())
.row(rowIndex)
.col(colIndex)
.header(cell.isHeaderCell())
.bBox(cell.getBBoxPdf())
.build();
page.getMainBody().add(tableCell);
TableCell tableCell = TableCell.builder().documentTree(context.getDocumentTree()).row(rowIndex).col(colIndex).header(cell.isHeaderCell()).bBox(cell.getBBoxPdf()).build();
List<Integer> treeId = context.getDocumentTree().createNewTableChildEntryAndReturnId(tableNode, tableCell);
tableCell.setTreeId(treeId);
@ -147,9 +120,7 @@ public class TableNodeFactory {
if (cell.getTextBlocks().isEmpty()) {
tableCell.setLeafTextBlock(context.getTextBlockFactory().emptyTextBlock(tableNode, context, page));
} else if (cell.getTextBlocks().size() == 1) {
textBlock = context.getTextBlockFactory()
.buildAtomicTextBlock2(cell.getTextBlocks()
.get(0).getSequences(), tableCell, context, page);
textBlock = context.getTextBlockFactory().buildAtomicTextBlock2(cell.getTextBlocks().get(0).getSequences(), tableCell, context, page);
tableCell.setLeafTextBlock(textBlock);
} else if (firstTextBlockIsHeadline(cell)) {
SectionNodeFactory.addSection(layoutParsingType,
@ -181,8 +152,7 @@ public class TableNodeFactory {
private boolean firstTextBlockIsHeadline(Cell cell) {
return cell.getTextBlocks()
.get(0).isHeadline();
return cell.getTextBlocks().get(0).isHeadline();
}
}

View File

@ -40,27 +40,26 @@ public class TextBlockFactory {
orientation = sequences.get(0).getDir().toString();
textRotation = sequences.get(0).getDir().getRotation();
}
return AtomicTextBlock.fromSearchTextWithTextPosition(searchTextWithTextPositionDto.getSearchText(),
searchTextWithTextPositionDto.getLineBreaks(),
searchTextWithTextPositionDto.getBoldTextBoundaries(),
searchTextWithTextPositionDto.getItalicTextBoundaries(),
searchTextWithTextPositionDto.getPositions(),
searchTextWithTextPositionDto.getStringIdxToPositionIdx(),
idx,
parent,
numberOnPage,
page,
offset,
orientation,
textRotation);
var atb = AtomicTextBlock.fromSearchTextWithTextPosition(searchTextWithTextPositionDto.getSearchText(),
searchTextWithTextPositionDto.getLineBreaks(),
searchTextWithTextPositionDto.getBoldTextBoundaries(),
searchTextWithTextPositionDto.getItalicTextBoundaries(),
searchTextWithTextPositionDto.getPositions(),
searchTextWithTextPositionDto.getStringIdxToPositionIdx(),
idx,
parent,
numberOnPage,
page,
offset,
orientation,
textRotation);
return atb;
}
public AtomicTextBlock emptyTextBlock(SemanticNode parent, DocumentGraphFactory.Context context, Page page) {
long idx = textBlockIdx;
textBlockIdx++;
return AtomicTextBlock.empty(idx, stringOffset, page, context.getAndIncrementTextBlockNumberOnPage(page), parent);
return emptyTextBlock(parent, context.getAndIncrementTextBlockNumberOnPage(page), page);
}
@ -68,7 +67,8 @@ public class TextBlockFactory {
long idx = textBlockIdx;
textBlockIdx++;
return AtomicTextBlock.empty(idx, stringOffset, page, numberOnPage, parent);
var atb = AtomicTextBlock.empty(idx, stringOffset, page, numberOnPage, parent);
return atb;
}
}

View File

@ -41,7 +41,9 @@ public class DocumentGraphMapper {
DocumentTree documentTree = new DocumentTree(document);
Context context = new Context(documentData, documentTree);
context.pages.addAll(Arrays.stream(documentData.getDocumentPages()).map(DocumentGraphMapper::buildPage).toList());
context.pages.addAll(Arrays.stream(documentData.getDocumentPages())
.map(DocumentGraphMapper::buildPage)
.toList());
context.documentTree.getRoot().getChildren().addAll(buildEntries(documentData.getDocumentStructure().getRoot().getChildren(), context));
@ -59,7 +61,9 @@ public class DocumentGraphMapper {
List<DocumentTree.Entry> newEntries = new LinkedList<>();
for (DocumentStructure.EntryData entryData : entries) {
List<Page> pages = Arrays.stream(entryData.getPageNumbers()).map(pageNumber -> getPage(pageNumber, context)).toList();
List<Page> pages = Arrays.stream(entryData.getPageNumbers())
.map(pageNumber -> getPage(pageNumber, context))
.toList();
SemanticNode node = switch (entryData.getType()) {
case SECTION -> buildSection(context);
@ -77,16 +81,17 @@ public class DocumentGraphMapper {
if (entryData.getAtomicBlockIds().length > 0) {
TextBlock textBlock = toTextBlock(entryData.getAtomicBlockIds(), context, node);
node.setLeafTextBlock(textBlock);
switch (entryData.getType()) {
case HEADER -> pages.forEach(page -> page.setHeader((Header) node));
case FOOTER -> pages.forEach(page -> page.setFooter((Footer) node));
default -> textBlock.getAtomicTextBlocks()
.forEach(atb -> atb.getPage().getTextBlocksOnPage().add(atb));
}
}
List<Integer> treeId = Arrays.stream(entryData.getTreeId()).boxed().toList();
List<Integer> treeId = Arrays.stream(entryData.getTreeId()).boxed()
.toList();
node.setTreeId(treeId);
switch (entryData.getType()) {
case HEADER -> pages.forEach(page -> page.setHeader((Header) node));
case FOOTER -> pages.forEach(page -> page.setFooter((Footer) node));
default -> pages.forEach(page -> page.getMainBody().add(node));
}
newEntries.add(DocumentTree.Entry.builder().treeId(treeId).children(buildEntries(entryData.getChildren(), context)).node(node).build());
}
return newEntries;
@ -142,6 +147,7 @@ public class DocumentGraphMapper {
return Section.builder().documentTree(context.documentTree).build();
}
private SuperSection buildSuperSection(Context context) {
return SuperSection.builder().documentTree(context.documentTree).build();
@ -166,22 +172,24 @@ public class DocumentGraphMapper {
private TextBlock toTextBlock(Long[] atomicTextBlockIds, Context context, SemanticNode parent) {
return Arrays.stream(atomicTextBlockIds).map(atomicTextBlockId -> getAtomicTextBlock(context, parent, atomicTextBlockId)).collect(new TextBlockCollector());
return Arrays.stream(atomicTextBlockIds)
.map(atomicTextBlockId -> getAtomicTextBlock(context, parent, atomicTextBlockId))
.collect(new TextBlockCollector());
}
private AtomicTextBlock getAtomicTextBlock(Context context, SemanticNode parent, Long atomicTextBlockId) {
return AtomicTextBlock.fromAtomicTextBlockData(context.documentTextDataBlockData.get(Math.toIntExact(atomicTextBlockId)),
context.atomicPositionBlockData.get(Math.toIntExact(atomicTextBlockId)),
parent,
getPage(context.documentTextDataBlockData.get(Math.toIntExact(atomicTextBlockId)).getPage(), context));
context.atomicPositionBlockData.get(Math.toIntExact(atomicTextBlockId)),
parent,
getPage(context.documentTextDataBlockData.get(Math.toIntExact(atomicTextBlockId)).getPage(), context));
}
private Page buildPage(DocumentPage p) {
return Page.builder().rotation(p.getRotation()).height(p.getHeight()).width(p.getWidth()).number(p.getNumber()).mainBody(new LinkedList<>()).build();
return Page.builder().rotation(p.getRotation()).height(p.getHeight()).width(p.getWidth()).number(p.getNumber()).textBlocksOnPage(new LinkedList<>()).build();
}
@ -206,8 +214,10 @@ public class DocumentGraphMapper {
this.documentTree = documentTree;
this.pages = new LinkedList<>();
this.documentTextDataBlockData = Arrays.stream(documentData.getDocumentTextData()).toList();
this.atomicPositionBlockData = Arrays.stream(documentData.getDocumentPositions()).toList();
this.documentTextDataBlockData = Arrays.stream(documentData.getDocumentTextData())
.toList();
this.atomicPositionBlockData = Arrays.stream(documentData.getDocumentPositions())
.toList();
}

View File

@ -1,9 +1,10 @@
package com.knecon.fforesight.service.layoutparser.processor.utils;
import java.util.Collection;
import java.util.Collections;
import java.util.Comparator;
import java.util.LinkedList;
import java.util.List;
import java.util.stream.Stream;
import com.knecon.fforesight.service.layoutparser.processor.model.AbstractPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.table.Cell;
@ -22,29 +23,83 @@ public class TableMergingUtility {
List<TablePageBlock> consecutiveTables = pageBlocks.stream()
.map(abstractPageBlock -> (TablePageBlock) abstractPageBlock)
.filter(tablePageBlock -> !tablePageBlock.equals(originalTablePageBlock))
.sorted(Comparator.comparingInt(TablePageBlock::getPage).thenComparing(TablePageBlock::getY).thenComparing(TablePageBlock::getX))
.toList();
assert consecutiveTables.size() == pageBlocks.size() - 1;
var currentTable = originalTablePageBlock;
int currentTableIndex = 0;
List<TablePageBlock> consecutiveTablesWithSameColCountAndHeaders = new LinkedList<>();
for (TablePageBlock consecutiveTable : consecutiveTables) {
if (consecutiveTable.getColCount() == originalTablePageBlock.getColCount() && !hasTableHeader(consecutiveTable) && outerBoundaryAlignsX(originalTablePageBlock,
consecutiveTable)) {
consecutiveTablesWithSameColCountAndHeaders.add(originalTablePageBlock);
for (int i = 0; i < consecutiveTables.size(); i++) {
TablePageBlock consecutiveTable = consecutiveTables.get(i);
if (consecutiveTable.getColCount() == originalTablePageBlock.getColCount() //
&& headersMatch(originalTablePageBlock, consecutiveTable) //
&& outerBoundaryAlignsX(originalTablePageBlock, consecutiveTable) //
&& consecutiveOrSamePage(currentTable, consecutiveTable) //
&& !tableBetween(currentTable, consecutiveTable, findTablesBetween(consecutiveTables, currentTableIndex, i))) {
currentTable = consecutiveTable;
currentTableIndex = i;
consecutiveTablesWithSameColCountAndHeaders.add(consecutiveTable);
}
}
return Stream.concat(Stream.of(originalTablePageBlock), consecutiveTablesWithSameColCountAndHeaders.stream()).toList();
return consecutiveTablesWithSameColCountAndHeaders;
}
private static List<TablePageBlock> findTablesBetween(List<TablePageBlock> consecutiveTables, int currentTableIndex, int i) {
if (currentTableIndex + 1 == consecutiveTables.size() || currentTableIndex + 1 >= i) {
return Collections.emptyList();
}
return consecutiveTables.subList(currentTableIndex + 1, i);
}
private static boolean consecutiveOrSamePage(TablePageBlock currentTable, TablePageBlock consecutiveTable) {
return currentTable.getPage() == consecutiveTable.getPage() || currentTable.getPage() + 1 == consecutiveTable.getPage();
}
private static boolean tableBetween(TablePageBlock currentTable, TablePageBlock consecutiveTable, List<TablePageBlock> tablesBetween) {
if (tablesBetween.isEmpty()) {
return false;
}
// assumes the tables are on the same page or on consecutive pages, all tables on pages in between are ignored.
return tablesBetween.stream()
.filter(tableBetween -> tableBetween.getPage() == currentTable.getPage())
.anyMatch(tableBetween -> tableBetween.isBelow(currentTable)) //
|| tablesBetween.stream()
.filter(tableBetween -> tableBetween.getPage() == consecutiveTable.getPage())
.anyMatch(tableBetween -> tableBetween.isAbove(consecutiveTable));
}
private static boolean headersMatch(TablePageBlock originalTable, TablePageBlock consecutiveTable) {
return getHeaders(consecutiveTable).isEmpty() || getHeaders(originalTable).equals(getHeaders(consecutiveTable));
}
private static boolean outerBoundaryAlignsX(TablePageBlock originalTablePageBlock, TablePageBlock consecutiveTable) {
return Math.abs(consecutiveTable.getMinX() - originalTablePageBlock.getMinX()) < TABLE_ALIGNMENT_THRESHOLD && Math.abs(consecutiveTable.getMaxX() - originalTablePageBlock.getMaxX()) < TABLE_ALIGNMENT_THRESHOLD;
return Math.abs(consecutiveTable.getMinX() - originalTablePageBlock.getMinX()) < TABLE_ALIGNMENT_THRESHOLD
&& Math.abs(consecutiveTable.getMaxX() - originalTablePageBlock.getMaxX()) < TABLE_ALIGNMENT_THRESHOLD;
}
private boolean hasTableHeader(TablePageBlock table) {
private List<Cell> getHeaders(TablePageBlock table) {
return table.getRows().stream().flatMap(Collection::stream).anyMatch(Cell::isHeaderCell);
return table.getRows()
.stream()
.flatMap(Collection::stream)
.filter(Cell::isHeaderCell)
.toList();
}
}

View File

@ -27,7 +27,7 @@ import lombok.extern.slf4j.Slf4j;
@Slf4j
public class LayoutparserEnd2EndTest extends AbstractTest {
public static final LayoutParsingType LAYOUT_PARSING_TYPE = LayoutParsingType.DOCUMINE;
public static final LayoutParsingType LAYOUT_PARSING_TYPE = LayoutParsingType.DOCUMINE_OLD;
@Autowired
private LayoutParsingPipeline layoutParsingPipeline;
@ -37,7 +37,7 @@ public class LayoutparserEnd2EndTest extends AbstractTest {
@Disabled
public void testLayoutParserEndToEnd() {
String filePath = "/home/kschuettler/Downloads/55974b3de7ed2915718a10458206bbd8.ORIGIN.pdf";
String filePath = "/home/kschuettler/Dokumente/Ticket Related/RED-9964/17a25133-e098-4610-b553-d1bf11a56d96/560e6ab1ab4754b9a62fd2e6d4d71327/560e6ab1ab4754b9a62fd2e6d4d71327.ORIGIN.pdf";
runForFile(filePath);
}