refactor: copy original pages in OCR instead of embedding rasterized PNGs

src/js/utils/ocr.ts

@@ -9,6 +9,7 @@ import {
   calculateSpaceTransform,
 } from './hocr-transform.js';
 import { getPDFDocument } from './helpers.js';
+import { loadPdfDocument } from './load-pdf-document.js';
 import { createConfiguredTesseractWorker } from './tesseract-runtime.js';
 
 export interface OcrOptions {
@@ -161,6 +162,7 @@ export async function performOcr(
     });
   }
 
+  const sourcePdfDoc = await loadPdfDocument(pdfBytes);
   const pdf = await getPDFDocument({ data: pdfBytes }).promise;
   const newPdfDoc = await PDFDocument.create();
 
@@ -255,44 +257,16 @@ export async function performOcr(
       );
       const data = result.data;
 
-      const newPage = newPdfDoc.addPage([viewport.width, viewport.height]);
-
-      const pngImageBytes = await new Promise<Uint8Array>(function (
-        resolve,
-        reject
-      ) {
-        canvas.toBlob(function (blob) {
-          if (!blob) {
-            reject(new Error('Failed to create image blob'));
-            return;
-          }
-          const reader = new FileReader();
-          reader.onload = function () {
-            resolve(new Uint8Array(reader.result as ArrayBuffer));
-          };
-          reader.onerror = function () {
-            reject(new Error('Failed to read image data'));
-          };
-          reader.readAsArrayBuffer(blob);
-        }, 'image/png');
-      });
-
-      // Release canvas memory
       canvas.width = 0;
       canvas.height = 0;
 
-      const pngImage = await newPdfDoc.embedPng(pngImageBytes);
+      const [copiedPage] = await newPdfDoc.copyPages(sourcePdfDoc, [i - 1]);
-      newPage.drawImage(pngImage, {
+      newPdfDoc.addPage(copiedPage);
-        x: 0,
-        y: 0,
-        width: viewport.width,
-        height: viewport.height,
-      });
 
       if (data.hocr) {
         const ocrPage = parseHocrDocument(data.hocr);
         drawOcrTextLayer(
-          newPage,
+          copiedPage,
           ocrPage,
           viewport.height,
           primaryFont,

src/tests/ocr.test.ts

@@ -3,11 +3,13 @@ import { beforeEach, describe, expect, it, vi } from 'vitest';
 const {
   createConfiguredTesseractWorker,
   getPDFDocument,
+  loadPdfDocument,
   getFontForLanguage,
   parseHocrDocument,
 } = vi.hoisted(() => ({
   createConfiguredTesseractWorker: vi.fn(),
   getPDFDocument: vi.fn(),
+  loadPdfDocument: vi.fn(),
   getFontForLanguage: vi.fn(),
   parseHocrDocument: vi.fn(),
 }));
@@ -25,13 +27,17 @@ const mockPdfPage = {
 
 const mockPdfOutputPage = {
   drawImage: vi.fn(),
+  drawPage: vi.fn(),
   drawText: vi.fn(),
+  setRotation: vi.fn(),
 };
 
 const mockPdfDoc = {
   registerFontkit: vi.fn(),
   embedFont: vi.fn(async () => ({ widthOfTextAtSize: vi.fn(() => 12) })),
   addPage: vi.fn(() => mockPdfOutputPage),
+  copyPages: vi.fn(async () => [mockPdfOutputPage]),
+  embedPage: vi.fn(async () => ({ width: 200, height: 100 })),
   embedPng: vi.fn(async () => ({ id: 'png' })),
   save: vi.fn(async () => new Uint8Array([1, 2, 3])),
 };
@@ -44,6 +50,10 @@ vi.mock('../js/utils/helpers.js', () => ({
   getPDFDocument,
 }));
 
+vi.mock('../js/utils/load-pdf-document.js', () => ({
+  loadPdfDocument,
+}));
+
 vi.mock('../js/utils/font-loader.js', () => ({
   getFontForLanguage,
 }));
@@ -77,6 +87,7 @@ describe('performOcr', () => {
   beforeEach(() => {
     createConfiguredTesseractWorker.mockReset();
     getPDFDocument.mockReset();
+    loadPdfDocument.mockReset();
     getFontForLanguage.mockReset();
     parseHocrDocument.mockReset();
 
@@ -90,10 +101,17 @@ describe('performOcr', () => {
     mockPdfDoc.registerFontkit.mockClear();
     mockPdfDoc.embedFont.mockClear();
     mockPdfDoc.addPage.mockClear();
+    mockPdfDoc.embedPage.mockClear();
     mockPdfDoc.embedPng.mockClear();
     mockPdfDoc.save.mockClear();
 
     createConfiguredTesseractWorker.mockResolvedValue(mockWorker);
+    loadPdfDocument.mockResolvedValue({
+      getPage: vi.fn(() => ({
+        width: 200,
+        height: 100,
+      })),
+    });
     getPDFDocument.mockReturnValue({
       promise: Promise.resolve({
         numPages: 1,