scribe.js-ocr 0.7.3 → 0.8.0

package/js/export/writePdf.js

@@ -10,6 +10,7 @@ import { createEmbeddedFontType0, createEmbeddedFontType1 } from './writePdfFont
 import { opt } from '../containers/app.js';
 import { pageMetricsArr } from '../containers/dataContainer.js';
 import ocr from '../objects/ocrObjects.js';
+import { getStyleLookup } from '../utils/miscUtils.js';
 
 /**
  * @param {number} x
@@ -97,6 +98,7 @@ export async function writePdf(hocrArr, minpage = 0, maxpage = -1, textMode = 'e
       normal: useOpt && FontCont.opt?.[familyKeyI]?.normal ? FontCont.opt[familyKeyI].normal : FontCont.raw[familyKeyI].normal,
       italic: useOpt && FontCont.opt?.[familyKeyI]?.italic ? FontCont.opt[familyKeyI].italic : FontCont.raw[familyKeyI].italic,
       bold: useOpt && FontCont.opt?.[familyKeyI]?.bold ? FontCont.opt[familyKeyI].bold : FontCont.raw[familyKeyI].bold,
+      boldItalic: useOpt && FontCont.opt?.[familyKeyI]?.boldItalic ? FontCont.opt[familyKeyI].boldItalic : FontCont.raw[familyKeyI].boldItalic,
     };
     await addFamilyObj(familyKeyI, familyObjI);
   }
@@ -301,6 +303,8 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
 
   const pdfFontsUsed = new Set();
 
+  const underlines = /** @type {Array<{left: number, right: number, top: number, height: number, fontSize: number, bold: boolean}>} */ ([]);
+
   // Start 1st object: Text Content
   let textContentObjStr = '';
 
@@ -349,7 +353,7 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
     let wordFontOpentype = (wordJ.lang === 'chi_sim' ? fontChiSim : wordFont.opentype);
 
     if (!wordFontOpentype) {
-      const fontNameMessage = wordJ.lang === 'chi_sim' ? 'chi_sim' : `${wordFont.family} (${wordJ.style})`;
+      const fontNameMessage = wordJ.lang === 'chi_sim' ? 'chi_sim' : `${wordFont.family} (${getStyleLookup(wordJ.style)})`;
       console.log(`Skipping word due to missing font (${fontNameMessage})`);
       continue;
     }
@@ -359,7 +363,7 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
     let wordFontSize = word0Metrics.fontSize;
 
     // Set font and font size
-    const pdfFontCurrent = wordJ.lang === 'chi_sim' ? pdfFonts.NotoSansSC.normal : pdfFonts[wordFont.family][wordJ.style];
+    const pdfFontCurrent = wordJ.lang === 'chi_sim' ? pdfFonts.NotoSansSC.normal : pdfFonts[wordFont.family][getStyleLookup(wordJ.style)];
     pdfFontNameCurrent = pdfFontCurrent.name;
     pdfFontTypeCurrent = pdfFontCurrent.type;
     pdfFontsUsed.add(pdfFontCurrent);
@@ -372,7 +376,7 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
     const word0LeftBearing = wordJ.visualCoords ? word0Metrics.leftSideBearing : 0;
 
     let tz = 100;
-    if (wordJ.dropcap) {
+    if (wordJ.style.dropcap) {
       const wordWidthActual = wordJ.bbox.right - wordJ.bbox.left;
       tz = (wordWidthActual / word0Metrics.visualWidth) * 100;
     }
@@ -406,6 +410,8 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
     let spacingAdj = 0;
     let kernSpacing = false;
     let wordLast = wordJ;
+    let underlineLeft = /** @type {?number} */ null;
+    let underlineRight = /** @type {?number} */ null;
     let wordFontOpentypeLast = wordFontOpentype;
     let fontSizeLast = wordFontSize;
     let tsCurrent = 0;
@@ -426,7 +432,7 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
       wordFontOpentype = wordJ.lang === 'chi_sim' ? fontChiSim : wordFont.opentype;
 
       if (!wordFontOpentype) {
-        const fontNameMessage = wordJ.lang === 'chi_sim' ? 'chi_sim' : `${wordFont.family} (${wordJ.style})`;
+        const fontNameMessage = wordJ.lang === 'chi_sim' ? 'chi_sim' : `${wordFont.family} (${getStyleLookup(wordJ.style)})`;
         console.log(`Skipping word due to missing font (${fontNameMessage})`);
         continue;
       }
@@ -446,11 +452,11 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
         fillColor = wordJ.matchTruth ? '0 1 0.5 rg' : '1 0 0 rg';
       }
 
-      const angleAdjWord = wordJ.sup ? ocr.calcWordAngleAdj(wordJ) : { x: 0, y: 0 };
+      const angleAdjWord = wordJ.style.sup ? ocr.calcWordAngleAdj(wordJ) : { x: 0, y: 0 };
       const angleAdjWordX = (rotateBackground && Math.abs(angle ?? 0) > 0.05) ? angleAdjWord.x : 0;
 
       let ts = 0;
-      if (wordJ.sup || wordJ.dropcap) {
+      if (wordJ.style.sup || wordJ.style.dropcap) {
         ts = (lineObj.bbox.bottom + lineObj.baseline[1] + angleAdjLine.y) - (wordJ.bbox.bottom + angleAdjLine.y + angleAdjWord.y);
         if (!wordJ.visualCoords) {
           const fontDesc = wordFont.opentype.descender / wordFont.opentype.unitsPerEm * wordMetrics.fontSize;
@@ -462,12 +468,12 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
 
       // TODO: This probably fails for Chinese, rethink.
       tz = 100;
-      if (wordJ.dropcap) {
+      if (wordJ.style.dropcap) {
         const wordWidthActual = wordJ.bbox.right - wordJ.bbox.left;
         tz = (wordWidthActual / wordMetrics.visualWidth) * 100;
       }
 
-      const pdfFont = wordJ.lang === 'chi_sim' ? pdfFonts.NotoSansSC.normal : pdfFonts[wordFont.family][wordJ.style];
+      const pdfFont = wordJ.lang === 'chi_sim' ? pdfFonts.NotoSansSC.normal : pdfFonts[wordFont.family][getStyleLookup(wordJ.style)];
       const pdfFontName = pdfFont.name;
       const pdfFontType = pdfFont.type;
       pdfFontsUsed.add(pdfFont);
@@ -480,7 +486,8 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
         // The space between words determined by:
         // (1) The right bearing of the last word, (2) the left bearing of the current word, (3) the width of the space character between words,
         // (4) the current character spacing value (applied twice--both before and after the space character).
-        const spaceWidthGlyph = wordFontOpentypeLast.charToGlyph(' ').advanceWidth * (fontSizeLast / wordFontOpentypeLast.unitsPerEm);
+        const spaceAdvance = wordFontOpentypeLast.charToGlyph(' ').advanceWidth || wordFontOpentypeLast.unitsPerEm / 2;
+        const spaceWidthGlyph = spaceAdvance * (fontSizeLast / wordFontOpentypeLast.unitsPerEm);
 
         const wordSpaceExpectedPx = (spaceWidthGlyph + charSpacingLast * 2 + wordRightBearingLast) + wordLeftBearing;
 
@@ -503,10 +510,11 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
       // However, this assumption does not hold for single-character words, as there is no space between character to adjust.
       // Therefore, we calculate the difference between the rendered and actual word and apply an adjustment to the width of the next space.
       // (This does not apply to drop caps as those have horizontal scaling applied to exactly match the image.)
-      if (charArr.length === 1 && !wordJ.dropcap) {
+      if (charArr.length === 1 && !wordJ.style.dropcap) {
         const wordLastGlyph = wordFontOpentype.charToGlyph(charArr.at(-1));
         const wordLastGlyphMetrics = wordLastGlyph.getMetrics();
-        const lastCharWidth = (wordLast.visualCoords ? (wordLastGlyphMetrics.xMax - wordLastGlyphMetrics.xMin) : wordLastGlyph.advanceWidth) * (wordFontSize / wordFontOpentype.unitsPerEm);
+        const lastCharAdvance = wordLast.visualCoords ? (wordLastGlyphMetrics.xMax - wordLastGlyphMetrics.xMin) : wordLastGlyph.advanceWidth || wordFontOpentype.unitsPerEm / 2;
+        const lastCharWidth = lastCharAdvance * (wordFontSize / wordFontOpentype.unitsPerEm);
         spacingAdj = wordWidthAdj - lastCharWidth - angleAdjWordX;
       } else {
         spacingAdj = 0 - angleAdjWordX;
@@ -514,7 +522,7 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
 
       textContentObjStr += ' ] TJ\n';
 
-      const fontSize = wordJ.smallCaps && wordJ.text[0] && wordJ.text[0] !== wordJ.text[0].toUpperCase() ? wordFontSize * wordFont.smallCapsMult : wordFontSize;
+      const fontSize = wordJ.style.smallCaps && wordJ.text[0] && wordJ.text[0] !== wordJ.text[0].toUpperCase() ? wordFontSize * wordFont.smallCapsMult : wordFontSize;
       if (pdfFontName !== pdfFontNameCurrent || fontSize !== fontSizeLast) {
         textContentObjStr += `${pdfFontName} ${String(fontSize)} Tf\n`;
         pdfFontNameCurrent = pdfFontName;
@@ -541,8 +549,8 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
       // Non-ASCII and special characters are encoded/escaped using winEncodingLookup
       for (let k = 0; k < charArr.length; k++) {
         const letterSrc = charArr[k];
-        const letter = wordJ.smallCaps ? charArr[k].toUpperCase() : charArr[k];
-        const fontSizeLetter = wordJ.smallCaps && letterSrc !== letter ? wordFontSize * wordFont.smallCapsMult : wordFontSize;
+        const letter = wordJ.style.smallCaps ? charArr[k].toUpperCase() : charArr[k];
+        const fontSizeLetter = wordJ.style.smallCaps && letterSrc !== letter ? wordFontSize * wordFont.smallCapsMult : wordFontSize;
 
         const letterEnc = pdfFontTypeCurrent === 0 ? wordFontOpentype.charToGlyphIndex(letter)?.toString(16).padStart(4, '0') : winEncodingLookup[letter];
         if (letterEnc) {
@@ -611,6 +619,28 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
         }
       }
 
+      if (wordJ.style.underline && underlineLeft === null) {
+        underlineLeft = wordJ.bbox.left;
+      }
+
+      if (wordJ.style.underline) {
+        underlineRight = wordJ.bbox.right;
+      }
+
+      if (underlineLeft !== null && (!wordJ.style.underline || j === words.length - 1)) {
+        underlines.push({
+          left: underlineLeft,
+          right: underlineRight,
+          top: lineTopAdj,
+          height: lineObj.bbox.bottom - lineObj.bbox.top,
+          fontSize: wordFontSize,
+          bold: wordJ.style.bold,
+        });
+
+        underlineLeft = null;
+        underlineRight = null;
+      }
+
       wordLast = wordJ;
       wordRightBearingLast = wordLast.visualCoords ? wordMetrics.rightSideBearing : 0;
       wordFontOpentypeLast = wordFontOpentype;
@@ -622,5 +652,13 @@ async function ocrPageToPDFStream(pageObj, outputDims, pdfFonts, textMode, angle
 
   textContentObjStr += 'ET';
 
+  // Add underlines
+  underlines.forEach((underline) => {
+    const underlineThickness = underline.bold ? Math.ceil(underline.fontSize / 12) : Math.ceil(underline.fontSize / 24);
+    const underlineOffset = Math.ceil(underline.fontSize / 12) + underlineThickness;
+
+    textContentObjStr += `\n${String(underline.left)} ${String(outputDims.height - underline.top - underlineOffset)} ${String(underline.right - underline.left)} ${underlineThickness} re\nf\n`;
+  });
+
   return { textContentObjStr, pdfFontsUsed };
 }

package/js/export/writePdfFonts.js

@@ -108,12 +108,12 @@ const generateFontFlags = (serif, italic, smallcap, symbolic) => { /* eslint-dis
  *
  * @param {opentype.Font} font - Opentype.js font object
  * @param {number} objIndex - Index for font descriptor PDF object
- * @param {('normal'|'italic'|'bold')} style - Style of the font
+ * @param {boolean} italic
  * @param {?number} embeddedObjIndex - Index for embedded font file PDF object.
  *  If not provided, the font will not be embedded in the PDF.
  * @returns {string} The font descriptor object string.
  */
-function createFontDescriptor(font, objIndex, style = 'normal', embeddedObjIndex = null) {
+function createFontDescriptor(font, objIndex, italic, embeddedObjIndex = null) {
   let objOut = `${String(objIndex)} 0 obj\n<</Type/FontDescriptor`;
 
   const namesTable = font.names.windows || font.names;
@@ -155,7 +155,7 @@ function createFontDescriptor(font, objIndex, style = 'normal', embeddedObjIndex
 
   // Symbolic is always set to false, even if the font contains glyphs outside the Adobe standard Latin character set.
   // This is because symbolic fonts are only used when embedded, and this does not appear to matter for embedded fonts.
-  objOut += `/Flags ${String(generateFontFlags(serif, style === 'italic', false, false))}`;
+  objOut += `/Flags ${String(generateFontFlags(serif, italic, false, false))}`;
 
   if (embeddedObjIndex === null || embeddedObjIndex === undefined) {
     objOut += '>>\nendobj\n\n';
@@ -175,12 +175,12 @@ function createFontDescriptor(font, objIndex, style = 'normal', embeddedObjIndex
  *
  * @param {opentype.Font} font - Opentype.js font object
  * @param {number} firstObjIndex - Index for the first PDF object
- * @param {('normal'|'italic'|'bold')} style - Style of the font
+ * @param {boolean} [italic=false] - Whether the font is italic.
  * @param {boolean} [isStandardFont=false] - Whether the font is a standard font.
  *  Standard fonts are not embedded in the PDF.
  * @returns {Array<string>}
  */
-export function createEmbeddedFontType1(font, firstObjIndex, style = 'normal', isStandardFont = false) {
+export function createEmbeddedFontType1(font, firstObjIndex, italic = false, isStandardFont = false) {
   // Start 1st object: Font Dictionary
   let fontDictObjStr = `${String(firstObjIndex)} 0 obj\n<</Type/Font/Subtype/Type1`;
 
@@ -193,7 +193,8 @@ export function createEmbeddedFontType1(font, firstObjIndex, style = 'normal', i
 
   fontDictObjStr += '/Widths[';
   for (let i = 0; i < win1252Chars.length; i++) {
-    const advanceNorm = Math.round(font.charToGlyph(win1252Chars[i]).advanceWidth * (1000 / font.unitsPerEm));
+    const advance = font.charToGlyph(win1252Chars[i]).advanceWidth || font.unitsPerEm;
+    const advanceNorm = Math.round(advance * (1000 / font.unitsPerEm));
     fontDictObjStr += `${String(advanceNorm)} `;
   }
   fontDictObjStr += ']/FirstChar 32/LastChar 255';
@@ -201,7 +202,7 @@ export function createEmbeddedFontType1(font, firstObjIndex, style = 'normal', i
   fontDictObjStr += `/FontDescriptor ${String(firstObjIndex + 1)} 0 R>>\nendobj\n\n`;
 
   // Start 2nd object: Font Descriptor
-  const fontDescObjStr = createFontDescriptor(font, firstObjIndex + 1, style, isStandardFont ? null : firstObjIndex + 2);
+  const fontDescObjStr = createFontDescriptor(font, firstObjIndex + 1, italic, isStandardFont ? null : firstObjIndex + 2);
 
   // objOut += `${String(firstObjIndex + 1)} 0 obj\n<</Type/FontDescriptor`;
 
@@ -249,13 +250,14 @@ export function createEmbeddedFontType1(font, firstObjIndex, style = 'normal', i
  *
  * @param {opentype.Font} font - Opentype.js font object
  * @param {number} firstObjIndex - Index for the first PDF object
+ * @param {boolean} [italic=false] - Whether the font is italic.
  *
  * This function does not produce "toUnicode" or "Widths" objects,
  * so any PDF it creates directly will lack usable copy/paste.
  * However, both of these objects will be created from the embedded file
  * when the result is run through mupdf.
  */
-export function createEmbeddedFontType0(font, firstObjIndex, style = 'normal') {
+export function createEmbeddedFontType0(font, firstObjIndex, italic = false) {
   // Start 1st object: Font Dictionary
   let fontDictObjStr = `${String(firstObjIndex)} 0 obj\n<</Type/Font/Subtype/Type0`;
 
@@ -282,7 +284,7 @@ export function createEmbeddedFontType0(font, firstObjIndex, style = 'normal') {
   toUnicodeStr += '\nendstream\nendobj\n\n';
 
   // Start 3rd object: FontDescriptor
-  const fontDescObjStr = createFontDescriptor(font, firstObjIndex + 1, style, firstObjIndex + 3);
+  const fontDescObjStr = createFontDescriptor(font, firstObjIndex + 1, italic, firstObjIndex + 3);
 
   // objOut += `${String(firstObjIndex + 2)} 0 obj\n`;
 

package/js/export/writeTabular.js

@@ -86,9 +86,9 @@ function createCellsSingle(ocrTableWords, extraCols = [], startRow = 0, xlsxMode
 
         if (xlsxMode) {
           let fontStyle;
-          if (wordObj.style === 'italic') {
+          if (wordObj.style.italic) {
             fontStyle = '<i/>';
-          } else if (wordObj.smallCaps) {
+          } else if (wordObj.style.smallCaps) {
             fontStyle = '<smallCaps/>';
           } else {
             fontStyle = '';

package/js/export/writeText.js

@@ -54,17 +54,21 @@ export function writeText(ocrCurrent, minpage = 0, maxpage = -1, reflowText = fa
 
         if (docxMode) {
           let fontStyle = '';
-          if (wordObj.style === 'italic') {
+          if (wordObj.style.italic) {
             fontStyle += '<w:i/>';
-          } else if (wordObj.style === 'bold') {
+          } else if (wordObj.style.bold) {
             fontStyle += '<w:b/>';
           }
 
-          if (wordObj.smallCaps) {
+          if (wordObj.style.smallCaps) {
             fontStyle += '<w:smallCaps/>';
           }
 
-          if (wordObj.sup) {
+          if (wordObj.style.underline) {
+            fontStyle += '<w:u w:val="single"/>';
+          }
+
+          if (wordObj.style.sup) {
             fontStyle += '<w:vertAlign w:val="superscript"/>';
           }
 
@@ -79,7 +83,7 @@ export function writeText(ocrCurrent, minpage = 0, maxpage = -1, reflowText = fa
             } else if (supPrev) {
               textStr = `${textStr}</w:t></w:r><w:r>${styleStr}<w:t xml:space="preserve"> `;
             // If this word is a superscript, no space is added between words.
-            } else if (wordObj.sup && i > 0) {
+            } else if (wordObj.style.sup && i > 0) {
               textStr = `${textStr}</w:t></w:r><w:r>${styleStr}<w:t xml:space="preserve">`;
             } else {
               textStr = `${textStr} </w:t></w:r><w:r>${styleStr}<w:t xml:space="preserve">`;
@@ -89,7 +93,7 @@ export function writeText(ocrCurrent, minpage = 0, maxpage = -1, reflowText = fa
           }
 
           fontStylePrev = fontStyle;
-          supPrev = wordObj.sup;
+          supPrev = wordObj.style.sup;
         } else if (newLine) {
           textStr = `${textStr}\n`;
         } else if (h > 0 || g > 0 || i > 0) {

package/js/extractTables.js

@@ -22,11 +22,11 @@ export function extractTableContent(pageObj, layoutObj) {
 
 // TODO: This currently creates junk rows with only punctuation, as those bounding boxes are so small they often do not overlap with other lines.
 /**
-   * Extracts words from a page that are within the bounding boxes of the table, organized into arrays of rows and columns.
-   * The output is in the form of a 3D array, where the first dimension is the row, the second dimension is the column, and the third dimension is the word.
-   * @param {OcrPage} pageObj
-   * @param {Array<import('./objects/layoutObjects.js').LayoutBoxBase>} boxes
-   */
+ * Extracts words from a page that are within the bounding boxes of the table, organized into arrays of rows and columns.
+ * The output is in the form of a 3D array, where the first dimension is the row, the second dimension is the column, and the third dimension is the word.
+ * @param {OcrPage} pageObj
+ * @param {Array<import('./objects/layoutObjects.js').LayoutBoxBase>} boxes
+ */
 export function extractSingleTableContent(pageObj, boxes) {
   /** @type {Array<OcrWord>} */
   const wordArr = [];