Better filter for bogus textboxes in pixFindBaselines()

* Remove bogus textblocks that are really just part of a real textblock
  but have very small height and are above or below the actual textblock.
* Continue to allow more than one textbox for each baseline.
  This is because large gaps between textblocks in a line make it
  difficult to join safely.
* Could also add a minimal vertical closing (c1.2) to filter in order to
  join bogus textboxes; not done yet because it may not be necessary.

prog/baseline_reg.c

@@ -131,6 +131,20 @@ L_REGPARAMS  *rp;
     pixDestroy(&pix2);
     pixDestroy(&pix5);
     numaDestroy(&na);
+    ptaDestroy(&pta);
+
+        /* Another test for baselines, with bogus short 'textblock' */
+    pixadb = pixaCreate(6);
+    pix1 = pixRead("baseline2.png");
+    na = pixFindBaselines(pix1, &pta, pixadb);
+    regTestCompareValues(rp, 3, numaGetCount(na), 0);  /* 11 */
+    pix2 = pixaDisplayTiledInRows(pixadb, 32, 1500, 1.0, 0, 30, 2);
+    regTestWritePixAndCheck(rp, pix2, IFF_PNG);  /* 12 */
+    pixDisplayWithTitle(pix2, 1400, 500, NULL, rp->display);
+    pixaDestroy(&pixadb);
+    pixDestroy(&pix1);
+    pixDestroy(&pix2);
+    numaDestroy(&na);
     ptaDestroy(&pta);
 
     return regTestCleanup(rp);

src/baseline.c

@@ -242,19 +242,27 @@ PTA       *pta;
         *ppta = pta;
     }
     if (pta) {
-      nloc = numaGetCount(naloc);
-      nbox = boxaGetCount(boxa3);
-      for (i = 0; i < nbox; i++) {
-          boxaGetBoxGeometry(boxa3, i, &bx, &by, &bw, &bh);
-          for (j = 0; j < nloc; j++) {
-              numaGetIValue(naloc, j, &locval);
-              if (L_ABS(locval - (by + bh)) > 25)
-                  continue;
-              ptaAddPt(pta, bx, locval);
-              ptaAddPt(pta, bx + bw, locval);
-              break;
-          }
-      }
+        nloc = numaGetCount(naloc);
+        nbox = boxaGetCount(boxa3);
+            /* For each textbox, find the corresponding baseline.
+             * There may be more than one textbox to a baseline.
+             * Bogus textboxes of very small height may have been
+             * generated, and these are removed.  Bogus textboxes can
+             * also be eliminated if the bottom is too far from any of
+             * the baselines.  Note that the boxes are an expansion from
+             * 4x reduction, so box parameters are multiples of 4. */
+        for (i = 0; i < nbox; i++) {
+            boxaGetBoxGeometry(boxa3, i, &bx, &by, &bw, &bh);
+            if (bh <= 8) continue;
+            for (j = 0; j < nloc; j++) {
+                numaGetIValue(naloc, j, &locval);
+                if (L_ABS(locval - (by + bh)) > 24)
+                    continue;
+                ptaAddPt(pta, bx, locval);
+                ptaAddPt(pta, bx + bw, locval);
+                break;
+            }
+        }
     }
     boxaDestroy(&boxa3);
 

version-notes.html

@@ -90,11 +90,15 @@ <h2 align=center>  <IMG SRC="moller52.jpg" border=1 ALIGN_MIDDLE> </h2>
 <pre>
 
 1.86.0  Not released
+        * Modify pixFindBaselines() to avoid joining textboxes and to
+          ignore bogus textboxes when listing baseline end points.
+        * Modify convertToPSEmbed() to efficiently encode webp input images.
         * Modify compressFilesToPdf() to allow upscale interpolation for
           low resolution pdfs.
-        * Source files changed: pageseg.c, pdfapp.c
-        * Prog files changed: binarizefiles.c, compresspdf.c, croppdf.c,
-          misctest2.c
+        * Source files changed: baseline.c, pageseg.c, pdfapp.c, psio1.c
+        * Prog files changed: baseline_reg.c, binarizefiles.c,
+          compresspdf.c, croppdf.c, misctest2.c,
+        * Prog files added: baseline2.png
 
 1.85.0  Oct 16, 2024
         * Use wrapper callSystemDebug() instead of system() in programs.