OCR reading of scanned PDFs — help
How it works
Reading happens in your browser, on your device: the PDF is not sent to any external OCR service. BackTax then proposes amounts that you check one by one; nothing is saved without your confirmation.
Recommended browsers
- Works best on: Chrome / Chromium ≥ 141 (≈ 1.2 s/page), Microsoft Edge ≥ 156 (≈ 1.2 s/page), Firefox ≥ 158 (≈ 1.3 s/page), WebKit (WebKitGTK, Linux) ≥ 2.52 (≈ 1.7 s/page)
- Safari (macOS, iPhone, iPad) and mobile browsers: not tested on a real device. WebKit was tested on Linux (Safari’s engine, not Safari itself).
- For long PDFs, use a computer rather than a phone and keep the tab open while reading.
Measured on 2026-10-25
Measurements per browser
| Browser | Verified version | Result | Text accuracy | Time per page | Peak memory | WASM SIMD · Workers · Threads |
|---|---|---|---|---|---|---|
| Chrome / Chromium | 141.0.7390.37 | Works best on | 99.8 % | 1.2 s/page | 837 MB | yes · yes · no |
| Microsoft Edge | 156.0.4314.42 | Works best on | 99.8 % | 1.2 s/page | 1371 MB | yes · yes · no |
| Firefox | 158.0.1 | Works best on | 99.8 % | 1.3 s/page | 1341 MB | yes · yes · no |
| WebKit (WebKitGTK, Linux) | 2.52.6 | Works best on | 99.8 % | 1.7 s/page | 1654 MB | yes · yes · no |
Measured on 2026-10-25 · Measurement environment : Linux x86_64 container, 4 vCPU, headless / Xvfb; Chromium and Edge via Playwright, Firefox via geckodriver, WebKitGTK MiniBrowser via WebKitWebDriver
Witness « BACKTAX OCR 48219 » + 2-page image documents in FR/DE/IT/EN. Accuracy = character-level similarity with the expected text.
Safari (macOS, iPhone, iPad) and mobile browsers: not tested on a real device. WebKit was tested on Linux (Safari’s engine, not Safari itself).
If reading does not work
You can always enter the amounts yourself while looking at the original PDF (Income → Add). No data is lost.