smart_document_capture_tesseract

Offline Tesseract engine for smart_document_capture, built on flutter_tesseract_ocr.

Works without internet, on Android and iOS.
Supports Persian (fas), English (eng) and every other Tesseract language.
Reads Tesseract's hOCR output, so every word has a box and a confidence — used by the core package for field scores and for drawing boxes in the review widget.

Setup

Add the packages:

dependencies:
  smart_document_capture: ^0.1.0
  smart_document_capture_tesseract: ^0.1.0

Download the trained data of your languages (for example fas.traineddata and eng.traineddata from tessdata_fast or the more accurate tessdata_best) into assets/tessdata/, and list them in assets/tessdata_config.json:
```
{ "files": ["fas.traineddata", "eng.traineddata"] }
```

Declare the assets in your app's pubspec.yaml:

flutter:
  assets:
    - assets/tessdata_config.json
    - assets/tessdata/

Usage

final capture = DocumentCapture(
  engine: TesseractOcrEngine(),
  options: const OcrOptions(languages: ['fas', 'eng']),
);

final result = await capture.process(
  DocumentImage.file(photo.path),
  ReceiptSchema.persian,
);

TesseractOcrEngine(args: {...}) sets Tesseract variables. The default page segmentation mode psm: 4 suits receipts; use 3 for full pages. Per-call variables can be passed in OcrOptions.extra.

Only DocumentImage.file is supported; save in-memory images to a file first.

Tips for better results

Photograph the document flat, well lit and filling the frame.
Crop and straighten the image before OCR (for example with a document scanner package).
tessdata_best is slower but noticeably better for Persian.

Parsing hOCR yourself

HocrParser.parse(hocr) turns any Tesseract hOCR output into an OcrResult, for example from a server-side Tesseract.

License

MIT