scribe.js-ocr 0.2.0 → 0.2.2

package/js/containers/imageContainer.js

@@ -254,7 +254,7 @@ export class ImageCache {
    * @returns
    */
   static #initMuPDFScheduler = async (numWorkers = 3) => {
-    const Tesseract = typeof process === 'undefined' ? (await import('../../tess/tesseract.esm.min.js')).default : await import('tesseract.js');
+    const Tesseract = typeof process === 'undefined' ? (await import('../../tess/tesseract.esm.min.js')).default : await import('@scribe.js/tesseract.js');
     const scheduler = await Tesseract.createScheduler();
     const workersPromiseArr = range(1, numWorkers).map(async () => {
       const w = await initMuPDFWorker();
@@ -300,7 +300,7 @@ export class ImageCache {
         page: n + 1, dpi, color, skipText: skipTextMode,
       }).then((res) => new ImageWrapper(n, res, color ? 'color' : 'gray'));
     }
-    throw new Error('No input mode set');
+    throw new Error('Attempted to render image without image input provided.');
   };
 
   /**

package/js/export/exportDebugCsv.js

@@ -22,10 +22,10 @@ const escapeCSVField = (field) => {
 };
 
 /**
-     * Converts an array of objects with atomic properties (string, number, boolean) to a CSV string.
-     * @param {Array<Object>} data - The array of data objects.
-     * @returns {string} - The CSV string.
-     */
+ * Converts an array of objects with atomic properties (string, number, boolean) to a CSV string.
+ * @param {Array<Object>} data - The array of data objects.
+ * @returns {string} - The CSV string.
+ */
 export const convertToCSV = (data) => {
   if (data.length === 0) {
     return '';

package/js/generalWorkerMain.js

@@ -198,7 +198,7 @@ export class gs {
       workerN = Math.min(Math.round((globalThis.navigator.hardwareConcurrency || 8) / 2), 6);
     }
 
-    const Tesseract = typeof process === 'undefined' ? (await import('../tess/tesseract.esm.min.js')).default : await import('tesseract.js');
+    const Tesseract = typeof process === 'undefined' ? (await import('../tess/tesseract.esm.min.js')).default : await import('@scribe.js/tesseract.js');
 
     gs.schedulerInner = await Tesseract.createScheduler();
     gs.schedulerInner.workers = new Array(workerN);

package/js/import/convertPageBlocks.js

@@ -9,7 +9,7 @@ import { getTextScript } from '../utils/miscUtils.js';
 
 /**
  * @param {Object} params
- * @param {Array<import('tesseract.js').Block>} params.ocrBlocks
+ * @param {Array<import('@scribe.js/tesseract.js').Block>} params.ocrBlocks
  * @param {number} params.n
  * @param {dims} params.pageDims
  * @param {number} params.rotateAngle - The angle that the input image is rotated prior to recognition.

package/js/import/import.js

@@ -465,6 +465,8 @@ export async function importFilesSupp(files, ocrName) {
 
   const ocrData = await importOCRFiles(ocrFilesAll);
 
+  const scribeMode = ocrData.scribeMode;
+
   const pageCountHOCR = ocrData.hocrRaw.length;
 
   // If both OCR data and image data are present, confirm they have the same number of pages
@@ -478,5 +480,5 @@ export async function importFilesSupp(files, ocrName) {
   if (ocrData.abbyyMode) format = 'abbyy';
   if (ocrData.stextMode) format = 'stext';
 
-  convertOCRAll(ocrData.hocrRaw, false, format, ocrName);
+  await convertOCRAll(ocrData.hocrRaw, false, format, ocrName, scribeMode);
 }

package/js/import/importOCR.js

@@ -18,13 +18,11 @@ export async function importOCRFiles(ocrFilesAll) {
   // In the case of 1 HOCR file
   const singleHOCRMode = ocrFilesAll.length === 1;
 
-  let hocrStrStart = '';
-  let hocrStrEnd = '';
+  let hocrStrStart = null;
   let abbyyMode = false;
   let stextMode = false;
   let scribeMode = false;
 
-  let hocrArrPages;
   let pageCountHOCR;
   let hocrRaw;
   /** @type  {?Object.<string, FontMetricsFamily>} */
@@ -47,20 +45,16 @@ export async function importOCRFiles(ocrFilesAll) {
     stextMode = !!node2 && !!/<document name/.test(node2);
 
     if (abbyyMode) {
-      hocrArrPages = hocrStrAll.split(/(?=<page)/).slice(1);
+      hocrRaw = hocrStrAll.split(/(?=<page)/).slice(1);
     } else if (stextMode) {
-      hocrArrPages = hocrStrAll.split(/(?=<page)/).slice(1);
+      hocrRaw = hocrStrAll.split(/(?=<page)/).slice(1);
     } else {
-      hocrStrStart = hocrStrAll.match(/[\s\S]*?<body>/)[0];
-      hocrStrEnd = hocrStrAll.match(/<\/body>[\s\S]*$/)[0];
-      hocrArrPages = splitHOCRStr(hocrStrAll);
+      // `hocrStrStart` will be missing for individual HOCR pages created with Tesseract.js or the Tesseract API.
+      hocrStrStart = hocrStrAll.match(/[\s\S]*?<body>/)?.[0];
+      hocrRaw = splitHOCRStr(hocrStrAll);
     }
 
-    pageCountHOCR = hocrArrPages.length;
-    hocrRaw = Array(pageCountHOCR);
-    for (let i = 0; i < pageCountHOCR; i++) {
-      hocrRaw[i] = hocrStrStart + hocrArrPages[i] + hocrStrEnd;
-    }
+    pageCountHOCR = hocrRaw.length;
   } else {
     pageCountHOCR = ocrFilesAll.length;
     hocrRaw = Array(pageCountHOCR);
@@ -76,11 +70,11 @@ export async function importOCRFiles(ocrFilesAll) {
     }
   }
 
-  if (!abbyyMode && !stextMode && hocrRaw[0]) {
+  if (!abbyyMode && !stextMode && hocrStrStart) {
     const getMeta = (name) => {
       const regex = new RegExp(`<meta name=["']${name}["'][^<]+`, 'i');
 
-      const nodeStr = hocrRaw[0].match(regex)?.[0];
+      const nodeStr = hocrStrStart.match(regex)?.[0];
       if (!nodeStr) return null;
       const contentStr = nodeStr.match(/content=["']([\s\S]+?)(?=["']\s{0,5}\/?>)/i)?.[1];
       if (!contentStr) return null;

package/js/recognizeConvert.js

@@ -24,7 +24,19 @@ import { replaceObjectProperties } from './utils/miscUtils.js';
  */
 export const compareOCRPage = async (pageA, pageB, options) => {
   const func = typeof process !== 'undefined' ? (await import('./worker/compareOCRModule.js')).compareOCRPageImp : gs.scheduler.compareOCRPageImp;
-  const binaryImage = await ImageCache.getBinary(pageA.n);
+
+  // Some combinations of options require the image to be provided, and some do not.
+  // We skip sending the image for those that do not, as in addition to helping performance,
+  // this is also necessary to run basic comparison scripts (e.g. benchmarking accuracy) without providing the image.
+  // TODO: Rework the options so this works better with types.
+  // At present TypeScript has no way of knowing that certain combinations of options go with each other.
+  const mode = options?.mode || 'stats';
+  const evalConflicts = options?.evalConflicts ?? true;
+  const supplementComp = options?.supplementComp ?? false;
+  const skipImage = (mode === 'stats' && !supplementComp) || (mode === 'comb' && !evalConflicts && !supplementComp);
+
+  const binaryImage = skipImage ? null : await ImageCache.getBinary(pageA.n);
+
   const pageMetricsObj = pageMetricsArr[pageA.n];
   return func({
     pageA, pageB, binaryImage, pageMetricsObj, options,
@@ -51,7 +63,7 @@ export const evalOCRPage = async (params) => {
  * Compare two sets of OCR data.
  * @param {Array<OcrPage>} ocrA
  * @param {Array<OcrPage>} ocrB
- * @param  {Parameters<import('./worker/compareOCRModule.js').compareOCRPageImp>[0]['options']} options
+ * @param  {Parameters<import('./worker/compareOCRModule.js').compareOCRPageImp>[0]['options']} [options]
  */
 export const compareOCR = async (ocrA, ocrB, options) => {
   /** @type {Parameters<typeof compareOCRPage>[2]} */

package/js/worker/compareOCRModule.js

@@ -486,10 +486,19 @@ async function penalizeWord(wordObjs) {
 export async function compareOCRPageImp({
   pageA, pageB, binaryImage, pageMetricsObj, options = {},
 }) {
-  const binaryImageBit = binaryImage.imageBitmap || await getImageBitmap(binaryImage.src);
-
-  const imageUpscaled = binaryImage.upscaled;
-  const imageRotated = binaryImage.rotated;
+  // The `binaryImage` argument is not sent for certain operations, which do not require it.
+  // For example, running a basic comparison between a page and the ground truth does not require having the image.
+  // The types do not currently reflect this, so this should be reworked at some point.
+  /** @type {?ImageBitmap} */
+  let binaryImageBit = null;
+  let imageUpscaled = false;
+  let imageRotated = false;
+
+  if (binaryImage) {
+    binaryImageBit = binaryImage.imageBitmap || await getImageBitmap(binaryImage.src);
+    imageUpscaled = binaryImage.upscaled;
+    imageRotated = binaryImage.rotated;
+  }
 
   const mode = options?.mode === undefined ? 'stats' : options?.mode;
   const editConf = options?.editConf === undefined ? false : options?.editConf;

package/js/worker/generalWorker.js

@@ -17,7 +17,7 @@ import { optimizeFont } from './optimizeFontModule.js';
 // import Tesseract from "../../tess/tesseract.esm.min.js";
 const browserMode = typeof process === 'undefined';
 
-const Tesseract = browserMode ? (await import('../../tess/tesseract.esm.min.js')).default : await import('tesseract.js/src/index.js');
+const Tesseract = browserMode ? (await import('../../tess/tesseract.esm.min.js')).default : await import('@scribe.js/tesseract.js');
 
 const defaultConfigs = {
   // TODO: Add back support for multiple PSM modes.
@@ -135,7 +135,7 @@ export const recognizeAndConvert = async ({
 
   const keepItalic = oemCurrent === 0;
 
-  const ocrBlocks = /** @type {Array<import('tesseract.js').Block>} */(res1.data.blocks);
+  const ocrBlocks = /** @type {Array<import('@scribe.js/tesseract.js').Block>} */(res1.data.blocks);
 
   const res2 = await convertPageBlocks({
     ocrBlocks, n, pageDims, rotateAngle: angle, keepItalic,
@@ -184,14 +184,14 @@ export const recognizeAndConvert2 = async ({
   let resLegacy;
   let resLSTM;
   if (options.lstm && options.legacy) {
-    const legacyBlocks = /** @type {Array<import('tesseract.js').Block>} */(res0.data.blocks);
+    const legacyBlocks = /** @type {Array<import('@scribe.js/tesseract.js').Block>} */(res0.data.blocks);
     resLegacy = await convertPageBlocks({
       ocrBlocks: legacyBlocks, n, pageDims, rotateAngle: angle, keepItalic: true, upscale: options.upscale,
     });
     (async () => {
       const res1 = await resArr[1];
 
-      const lstmBlocks = /** @type {Array<import('tesseract.js').Block>} */(res1.data.blocks);
+      const lstmBlocks = /** @type {Array<import('@scribe.js/tesseract.js').Block>} */(res1.data.blocks);
       resLSTM = await convertPageBlocks({
         ocrBlocks: lstmBlocks, n, pageDims, rotateAngle: angle, keepItalic: false, upscale: options.upscale,
       });
@@ -201,12 +201,12 @@ export const recognizeAndConvert2 = async ({
       postMessage({ data: xB, id: `${id}b` });
     })();
   } else if (!options.lstm && options.legacy) {
-    const legacyBlocks = /** @type {Array<import('tesseract.js').Block>} */(res0.data.blocks);
+    const legacyBlocks = /** @type {Array<import('@scribe.js/tesseract.js').Block>} */(res0.data.blocks);
     resLegacy = await convertPageBlocks({
       ocrBlocks: legacyBlocks, n, pageDims, rotateAngle: angle, keepItalic: true, upscale: options.upscale,
     });
   } else if (options.lstm && !options.legacy) {
-    const lstmBlocks = /** @type {Array<import('tesseract.js').Block>} */(res0.data.blocks);
+    const lstmBlocks = /** @type {Array<import('@scribe.js/tesseract.js').Block>} */(res0.data.blocks);
     resLSTM = await convertPageBlocks({
       ocrBlocks: lstmBlocks, n, pageDims, rotateAngle: angle, keepItalic: false, upscale: options.upscale,
     });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "scribe.js-ocr",
-  "version": "0.2.0",
+  "version": "0.2.2",
   "description": "High-quality OCR and text extraction for images and PDFs.",
   "main": "scribe.js",
   "directories": {
@@ -52,7 +52,7 @@
     "canvas": "^2.11.2",
     "commander": "^11.1.0",
     "puppeteer": "^22.13.0",
-    "tesseract.js": "scribeocr/tesseract.js#2065fd6",
+    "@scribe.js/tesseract.js": "^5.0.5",
     "web-worker": "~1.2.0"
   }
 }

package/scribe.js

@@ -11,7 +11,7 @@ import { ImageCache } from './js/containers/imageContainer.js';
 import coords from './js/coordinates.js';
 import { drawDebugImages } from './js/debug.js';
 import { download, exportData } from './js/export/export.js';
-import { writeDebugCsv } from './js/export/exportDebugCsv.js';
+import { writeDebugCsv, convertToCSV } from './js/export/exportDebugCsv.js';
 import { extractSingleTableContent } from './js/export/exportWriteTabular.js';
 import { loadBuiltInFontsRaw, enableFontOpt } from './js/fontContainerMain.js';
 import { gs } from './js/generalWorkerMain.js';
@@ -131,6 +131,8 @@ class utils {
   // Misc utils
   static calcBoxOverlap = calcBoxOverlap;
 
+  static convertToCSV = convertToCSV;
+
   static replaceSmartQuotes = replaceSmartQuotes;
 
   static getRandomAlphanum = getRandomAlphanum;