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

RED-9974: Improved headline detection for documine old

Browse Source
This commit is contained in:
Dominique Eifländer 2024-08-30 09:43:31 +02:00
parent e3e9d16145
commit bb40345f79
3 changed files with 112 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/LayoutParsingPipeline.java
View File

@ -119,14 +119,18 @@ public class LayoutParsingPipeline {
log.info("Starting layout parsing for {}", layoutParsingRequest.identifier());
File originFile = layoutParsingStorageService.getOriginFile(layoutParsingRequest.originFileStorageId());
File viewerDocumentFile = layoutParsingStorageService.getViewerDocFile(layoutParsingRequest.viewerDocumentStorageId()).orElse(originFile);
File viewerDocumentFile = layoutParsingStorageService.getViewerDocFile(layoutParsingRequest.viewerDocumentStorageId())
.orElse(originFile);
VisualLayoutParsingResponse visualLayoutParsingResponse = layoutParsingRequest.visualLayoutParsingFileId()
.map(layoutParsingStorageService::getVisualLayoutParsingFile).orElse(new VisualLayoutParsingResponse());
.map(layoutParsingStorageService::getVisualLayoutParsingFile)
.orElse(new VisualLayoutParsingResponse());
ImageServiceResponse imageServiceResponse = layoutParsingRequest.imagesFileStorageId()
.map(layoutParsingStorageService::getImagesFile).orElse(new ImageServiceResponse());
.map(layoutParsingStorageService::getImagesFile)
.orElse(new ImageServiceResponse());
TableServiceResponse tableServiceResponse = layoutParsingRequest.tablesFileStorageId()
.map(layoutParsingStorageService::getTablesFile).orElse(new TableServiceResponse());
.map(layoutParsingStorageService::getTablesFile)
.orElse(new TableServiceResponse());
ClassificationDocument classificationDocument = parseLayout(settings.getLayoutParsingTypeOverride() == null //
? layoutParsingRequest.layoutParsingType() : settings.getLayoutParsingTypeOverride(),
@ -143,13 +147,20 @@ public class LayoutParsingPipeline {
log.info("Creating viewer document for {}", layoutParsingRequest.identifier());
layoutGridService.addLayoutGrid(viewerDocumentFile, documentGraph, viewerDocumentFile, false, layoutParsingRequest.visualLayoutParsingFileId().isPresent());
layoutGridService.addLayoutGrid(viewerDocumentFile,
documentGraph,
viewerDocumentFile,
false,
layoutParsingRequest.visualLayoutParsingFileId()
.isPresent());
log.info("Storing resulting files for {}", layoutParsingRequest.identifier());
layoutParsingStorageService.storeDocumentData(layoutParsingRequest, DocumentDataMapper.toDocumentData(documentGraph));
if (layoutParsingRequest.documentMarkdownFileStorageId().isPresent()) {
layoutParsingStorageService.storeMarkdownFile(layoutParsingRequest.documentMarkdownFileStorageId().get(), new MarkdownMapper().toMarkdownContent(documentGraph));
if (layoutParsingRequest.documentMarkdownFileStorageId()
.isPresent()) {
layoutParsingStorageService.storeMarkdownFile(layoutParsingRequest.documentMarkdownFileStorageId()
.get(), new MarkdownMapper().toMarkdownContent(documentGraph));
}
layoutParsingStorageService.storeSimplifiedText(layoutParsingRequest, simplifiedSectionTextService.toSimplifiedText(documentGraph));
layoutParsingStorageService.storeViewerDocument(layoutParsingRequest, viewerDocumentFile);
@ -379,6 +390,12 @@ public class LayoutParsingPipeline {
case CLARIFYND -> clarifyndClassificationService.classifyDocument(classificationDocument);
}
if (layoutParsingType.equals(LayoutParsingType.DOCUMINE_OLD)) {
for (ClassificationPage page : classificationDocument.getPages()) {
docuMineBlockificationService.mergeblocks(page, page.getCleanRulings().withoutTextRulings(), 0, 10);
}
}
List<TextPageBlock> headlines = classificationDocument.getPages()
.stream()
.flatMap(classificationPage -> classificationPage.getTextBlocks()

83
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/blockification/DocuMineBlockificationService.java
View File

@ -2,19 +2,23 @@ package com.knecon.fforesight.service.layoutparser.processor.services.blockifica
import java.util.ArrayList;
import java.util.List;
import java.util.ListIterator;
import java.util.regex.Matcher;
import java.util.regex.Pattern;
import java.util.stream.Collectors;
import org.springframework.stereotype.Service;
import com.knecon.fforesight.service.layoutparser.internal.api.data.redaction.LayoutEngine;
import com.knecon.fforesight.service.layoutparser.processor.model.AbstractPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.ClassificationPage;
import com.knecon.fforesight.service.layoutparser.processor.model.Orientation;
import com.knecon.fforesight.service.layoutparser.processor.model.table.CleanRulings;
import com.knecon.fforesight.service.layoutparser.processor.model.table.TablePageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPageBlock;
import com.knecon.fforesight.service.layoutparser.processor.model.text.TextPositionSequence;
@SuppressWarnings("all")
@Service
public class DocuMineBlockificationService {
@ -57,8 +61,11 @@ public class DocuMineBlockificationService {
boolean isSplitByRuling = prev != null && usedRulings.lineBetween(prev, word);
boolean splitByDir = prev != null && !prev.getDir().equals(word.getDir());
boolean splitByOtherFontAndOtherY = prev != null && Math.abs(prev.getMaxYDirAdj() - word.getMaxYDirAdj()) > word.getTextHeight() * 0.2 //
&& (word.getFontStyle().contains("bold") && !prev.getFontStyle().contains("bold") //
|| prev.getFontStyle().contains("bold") && !word.getFontStyle().contains("bold"));
&& (word.getFontStyle().contains("bold") && !prev.getFontStyle().contains("bold")
//
|| prev.getFontStyle().contains("bold") && !word.getFontStyle().contains("bold")
|| Math.abs(prev.getFontSize() - word.getFontSize()) >= 1
|| Math.abs(word.getTextHeight() - prev.getTextHeight()) > 0.8);
Matcher matcher = pattern.matcher(chunkWords.stream()
.collect(Collectors.joining(" ")).toString());
@ -120,5 +127,77 @@ public class DocuMineBlockificationService {
return new ClassificationPage(textPageBlocks);
}
public void mergeblocks(ClassificationPage page, CleanRulings usedRulings, float xThreshold, float yThreshold) {
var blocks = page.getTextBlocks();
ListIterator<AbstractPageBlock> itty = blocks.listIterator();
while (itty.hasNext()) {
AbstractPageBlock block = itty.next();
if (block == null) {
continue;
}
if (block instanceof TablePageBlock) {
continue;
}
TextPageBlock current = (TextPageBlock) block;
for (int i = 0; i < blocks.size(); i++) {
AbstractPageBlock abstractPageBlock = blocks.get(i);
if (abstractPageBlock == null) {
continue;
}
if (abstractPageBlock == current) {
continue;
}
if (abstractPageBlock instanceof TablePageBlock) {
continue;
}
if (isHeadlineFromOutline(current) || isHeadlineFromOutline(abstractPageBlock)) {
continue;
}
TextPageBlock inner = (TextPageBlock) abstractPageBlock;
if (usedRulings.lineBetween(current, blocks.get(i))) {
continue;
}
if (current.getDir() == inner.getDir() && current.intersects(inner, yThreshold, xThreshold) && (current.getClassification() == null || current.getClassification()
.equals(inner.getClassification()))) {
boolean toDuplicate = current.isToDuplicate() || inner.isToDuplicate();
current.getSequences().addAll(inner.getSequences());
current = buildTextBlock(current.getSequences(), 0);
current.setClassification(inner.getClassification());
current.setToDuplicate(toDuplicate);
blocks.set(i, null);
itty.set(current);
}
}
}
var blocksIterator = blocks.iterator();
while (blocksIterator.hasNext()) {
if (blocksIterator.next() == null) {
blocksIterator.remove();
}
}
}
private boolean isHeadlineFromOutline(AbstractPageBlock abstractPageBlock) {
return abstractPageBlock.getEngines().contains(LayoutEngine.OUTLINE) && abstractPageBlock.getClassification() != null && abstractPageBlock.getClassification().isHeadline();
}
public static TextPageBlock buildTextBlock(List<TextPositionSequence> wordBlockList, int indexOnPage) {
return new TextPageBlock(wordBlockList);
}
}

11
layoutparser-service/layoutparser-service-processor/src/main/java/com/knecon/fforesight/service/layoutparser/processor/services/classification/DocuMineClassificationService.java
View File

@ -74,7 +74,7 @@ public class DocuMineClassificationService {
return;
}
if (document.getFontSizeCounter().getMostPopular() == null) {
textBlock.setClassification(PageBlockType.OTHER);
textBlock.setClassification(PageBlockType.PARAGRAPH);
return;
}
if (MarkedContentUtils.intersects(textBlock, page.getMarkedContentBboxPerType(), MarkedContentUtils.HEADER) //
@ -108,7 +108,10 @@ public class DocuMineClassificationService {
&& Character.isDigit(textBlock.toString().charAt(0))
&& atLeast3Matcher.reset().find()
&& !textBlock.toString().contains(":") //
|| textBlock.toString().equals(textBlock.toString().toUpperCase(Locale.ROOT)) && atLeast3Matcher.reset().find() && !textBlock.toString().contains(":") //
|| textBlock.toString().equals(textBlock.toString().toUpperCase(Locale.ROOT))
&& atLeast3Matcher.reset().find()
&& !textBlock.toString().contains(":")
&& !textBlock.toString().startsWith("(")//
|| textBlock.toString().startsWith("APPENDIX") //
|| textBlock.toString().startsWith("FIGURE") //
|| textBlock.toString().startsWith("Continued TABLE") //
@ -143,9 +146,9 @@ public class DocuMineClassificationService {
&& PositionUtils.getApproxLineCount(textBlock) < 2.9) {
textBlock.setClassification(PageBlockType.PARAGRAPH_ITALIC);
} else if (PositionUtils.isWithinBodyTextFrame(bodyTextFrame, textBlock)) {
textBlock.setClassification(PageBlockType.PARAGRAPH_UNKNOWN);
textBlock.setClassification(PageBlockType.PARAGRAPH);
} else {
textBlock.setClassification(PageBlockType.OTHER);
textBlock.setClassification(PageBlockType.PARAGRAPH);
}
}