Overview
Receipts and invoices pile up as photos and scans that nobody can search, total, or paste into a spreadsheet. Running OCR turns those images into selectable text so you can copy line items, amounts, dates, and vendor names into your bookkeeping in seconds. Clear Canvas runs OCR with tesseract.js (WebAssembly) entirely in your browser, which matters for finance documents: account numbers and totals never leave your device, and only the language model file is downloaded on first use.
Example workflow
- 1
Capture a clean image
Photograph or scan the receipt flat, well lit, and square to the camera. Clean dark text on a white background reaches the highest confidence.
- 2
Pick the document language
Select the matching OCR language — English, Chinese (Traditional or Simplified), Japanese, Korean, Spanish, German, French, or Portuguese — so characters are recognized correctly.
- 3
Run OCR on-device
Extract the text in the browser. Printed receipts on white paper typically reach 90%+ confidence.
- 4
Spot-check the numbers
Verify totals, tax, and dates against the image — these fields matter most for bookkeeping and are the most worth a quick human check.
- 5
Paste into your ledger
Copy the recognized text into your spreadsheet, accounting tool, or expense report.
Best settings
- Shoot flat, evenly lit, and cropped tight to the receipt; clean text on white yields the documented 90%+ confidence.
- Always set the OCR language to match the document rather than relying on a default.
- Increase capture resolution for faint thermal-printer receipts before running OCR.
- De-skew and crop out background clutter so the engine focuses on the text block.
When this works well
- Monthly expense and reimbursement workflows that start from phone photos of receipts.
- Pulling line items and totals from printed invoices into a spreadsheet.
- Privacy-sensitive finance documents that must not be uploaded to a third-party OCR service.
When to use another workflow
- Handwritten receipts and notes — accuracy drops on handwriting, so expect manual correction.
- Complex multi-column invoice layouts may need manual cleanup after extraction.
Common mistakes
- Photographing receipts at an angle or in shadow, which lowers recognition confidence.
- Leaving the language on the wrong setting for non-Latin scripts.
- Trusting extracted totals without a quick visual check against the original.
Frequently asked questions
How accurate is OCR on receipts?
Clean printed receipts on a white background typically reach 90%+ confidence. Faded thermal paper, angled photos, and handwriting reduce accuracy, so always confirm totals and dates.
Which languages can it read on invoices?
English, Chinese (Traditional and Simplified), Japanese, Korean, Spanish, German, French, and Portuguese. Set the language to match the document for best results.
Are my financial documents uploaded?
No. OCR runs in your browser via tesseract.js (WebAssembly); only the language model downloads on first use, and the receipt image itself never leaves your device.
Related tools and workflows
Tester ce cas d'usage maintenant
Ouvrez OCR d'image et appliquez ce workflow dans votre navigateur.