fix: Merge only blocks of text with the same style into the same paragraph.

Author: hnc-leebd

java/opendataloader-pdf-core/src/main/java/org/opendataloader/pdf/processors/ParagraphProcessor.java

@@ -16,6 +16,7 @@
 import org.verapdf.wcag.algorithms.entities.enums.TextAlignment;
 import org.verapdf.wcag.algorithms.semanticalgorithms.utils.CaptionUtils;
 import org.verapdf.wcag.algorithms.semanticalgorithms.utils.ChunksMergeUtils;
+import org.verapdf.wcag.algorithms.semanticalgorithms.utils.NodeUtils;
 
 import java.util.ArrayList;
 import java.util.Iterator;
@@ -302,6 +303,9 @@ private static List<TextBlock> processOtherLines(List<TextBlock> textBlocks) {
     }
 
     private static boolean isOneParagraph(TextBlock previousBlock, TextBlock nextBlock) {
+        if (!areCloseStyle(previousBlock, nextBlock)) {
+            return false;
+        }
         double probability = getDifferentLinesProbability(previousBlock, nextBlock);
         return CaptionUtils.areOverlapping(previousBlock.getLastLine(), nextBlock.getFirstLine().getBoundingBox()) &&
                 probability > DIFFERENT_LINES_PROBABILITY &&
@@ -348,4 +352,10 @@ private static double getDifferentLinesProbability(TextBlock previousBlock, Text
         }
         return 0;
     }
+
+    private static boolean areCloseStyle(TextBlock previousBlock, TextBlock nextBlock) {
+        return NodeUtils.areCloseNumbers(previousBlock.getFontSize(), nextBlock.getFontSize(), 1e-1) &&
+            NodeUtils.areCloseNumbers(previousBlock.getFirstLine().getFirstTextChunk().getFontWeight(),
+                nextBlock.getFirstLine().getFirstTextChunk().getFontWeight(), 1e-1);
+    }
 }

resources/1901.03003.html

@@ -15,7 +15,9 @@ <h3>Abstract</h3>
 
 <p>Keywords: Scene text recognition, optical character recognition, deep learning.</p>
 
-<p>∗Corresponding author 1https://github.com/Canjie-Luo/MORAN_v2</p>
+<p>∗Corresponding author</p>
+
+<p>1https://github.com/Canjie-Luo/MORAN_v2</p>
 
 <img src="1901.03003_figures/figure1.png" alt="figure1">