Bengali OCR: the free Tesseract model that reads it best
93.2%
of letters, with Tesseract's ben best model
Five free models, close together
We ran Tesseract 5.5.4 on 29 proofread pages from 10 printed books on bn.wikisource.
| Model | Letters | Words |
|---|---|---|
| ben best | 93.2% | 75.9% |
| script/Bengali | 92.6% | 74.5% |
| script/Bengali best | 92.4% | 74.1% |
| ben, tessdata repo | 92.0% | 74.5% |
| ben | 91.7% | 74.0% |
| ben+eng | 85.2% | 72.3% |
ben best came first on 6 of 10 books.
ben best is slower: 4.5 s a page, against 2.0 s for ben.
Letters are proofread letters found in order. Words are whole words right, which is what a search needs.
The command: one model file
curl -LO https://raw.githubusercontent.com/tesseract-ocr/tessdata_best/4.1.0/ben.traineddata
mkdir -p tessdata && mv ben.traineddata tessdata/
tesseract page.png page --tessdata-dir tessdata -l ben
It writes page.txt.
Homebrew's brew install tesseract-lang has the fast
ben. Use -l ben (91.7% here).
A scanned PDF: OCRmyPDF
cp -r "$(brew --prefix)/share/tessdata/configs" tessdata/
TESSDATA_PREFIX=$PWD/tessdata ocrmypdf -l ben --jobs 2 \
--sidecar book.txt scan.pdf book.pdf
On a 29-page PDF of the same scans it found 93.3% of letters in 119 s. It wrote book.txt and a searchable PDF/A.
Skip the first line and OCRmyPDF stops with
TesseractConfigError. Tesseract looks for its
configs/hocr file next to the model.
Three things that tripped us
- ben+eng on pages with no English wrote 1,621 Latin letters. Below, it turned the name মাধাই into ates.
- The script model also reads Assamese. script/Bengali wrote the Assamese ra ৰ 66 times, where Bengali has র. The proofread text has ৰ twice.
-
The word-list flags
-c load_system_dawg=0 -c load_freq_dawg=0changed nothing for ben best on all 29 pages, as in our dictionary test.
| Printed | “আয়রে মাধাই! কাছে আয়, |
| ben best | “ আয়রে মাধাই | কাছে আয়, |
| ben | «আয়রে মাধাষ্ট ! কাছে আর, |
| ben+eng | «আয়রে ates ! কাছে আর, |
Apple Vision on macOS 26.6 lists 30 languages. Bengali is not one of them.
How we ran it
- Pages marked proofread or validated on bn.wikisource, from a seeded shuffle of books, 3 a book (one book had 2 that fit). Pages with footnotes, tables or running heads in the text were left out. Verse was kept.
- 26 pages came from PDFs, 3 from a DjVu file, 1,036 to 1,920 px wide.
- Models from tessdata_fast, tessdata_best and tessdata at tag 4.1.0. The tessdata repo's ben is a different file (1.8 MB), so we ran it too.
- One thread a page, 2 pages at a time, at low priority, while other jobs ran. Times are rough.
- OCRmyPDF 17.13.0. The PDF holds the 29 scans as JPEGs, one a page.
- Licenses: Tesseract and tessdata are Apache-2.0, OCRmyPDF is MPL-2.0. All 10 scan files are marked public domain: 8 on Commons, 2 on bn.wikisource. We show no page images.
Scanned pages to text, no install
Run it yourself. Our Apify Actor turns a scanned PDF into Markdown or text. $0.006 a scanned page. It reads 18 languages, not Bengali yet.
Or send us the scan. Tack, our AI, does the job for $29, up to 50 pages. First page free. For Bengali, ask first.
Questions people ask
Does Tesseract support Bengali?
Yes, with the ben model. On our 29 pages, ben from tessdata_best read 93.2% of letters, the fast ben 91.7%.
Is Bangla OCR different from Bengali OCR?
No. Bangla is the language's own name for Bengali. Tesseract reads both with the same model, ben.
Is there a free Bengali OCR?
Tesseract and its models are free (Apache-2.0) and run offline. OCRmyPDF (MPL-2.0) adds a text layer to a scanned PDF.
Can I OCR a Bengali PDF?
Yes. OCRmyPDF with -l ben and the ben best model found 93.3% of letters on our 29-page PDF and wrote a searchable PDF/A.
Should I use ben+eng?
Only when the page has English on it. Our pages had none, and ben+eng found 85.2% of letters against 91.7% for ben alone.
Related guides
- Tamil OCR, tested: free Tesseract models on 30 scans
- Sanskrit OCR, tested: free Tesseract models on 30 scans
- Tesseract Hindi OCR: hin fast vs best vs hin+eng, tested
- tessdata_best vs tessdata_fast, tested on real scans
- How to make a PDF searchable: scanned PDFs, tested
- Odia OCR, tested: free Tesseract models on 30 scans
- Assamese OCR, tested: free Tesseract models on 30 scans