Guide
This is a free business card OCR that runs entirely in your browser: drop a photo of one card and you get the image automatically cropped to the card frame with perspective correction, the full text, and any email address, phone number, URL, and postal code found on it. Nothing is uploaded — the OCR engine (PaddleOCR compiled to WebAssembly) runs on your device.
How to OCR a business card
- Drop a card photo onto the tool (or click to choose a file). PNG, JPEG, and WebP work.
- Wait a moment — the first run downloads the OCR model (~25MB), which is cached afterwards.
- Get the cropped card image and the recognized text. Email, phone, URL, and postal code are highlighted separately.
- Copy the text or download the cropped image.
What you get
| Output | Detail |
|---|---|
| Cropped image | The card's edges are detected and the photo is warped flat (perspective-corrected, like a scanner app); when the outline can't be found, it falls back to cropping around the text. Downloadable as JPEG |
| Full text | Every line the OCR could read, in reading order |
| Contact hints | Email / phone / URL / postal code, extracted with pattern matching |
Example
A photo of a card reading 「株式会社サンプル 鈴木 良介 / [email protected] / 03-1234-5678」 on a desk returns the card neatly cropped out of the desk background, the full text, and [email protected] / 03-1234-5678 as ready-to-copy chips.
When to use the batch AI tool instead
This tool reads one card at a time and does not split names, companies, or titles into fields. If you have a stack of cards from an event and want name / company / title / email / phone / address in separate columns — exported as a CRM-ready CSV or a vCard file, with run history — use Convert business cards to CSV.
Limitations
- Works best on flat, well-lit, straight-on photos; strong perspective or glare reduces accuracy.
- Handwriting and heavily stylized logo type are often misread.
- Edge detection needs some contrast between card and background — a white card on a white desk falls back to the looser text-based crop, without perspective correction.
