What OCR can and can't do
Printed text in any major language — almost perfect accuracy on a clean, well-lit photo.
Screenshots and PDFs — essentially 100% accuracy because the rendering is already digital.
Handwriting — accurate for neat block letters, less so for cursive. Doctors' notes are still the final boss.
Tables, math equations, music notation — partial support; expect manual cleanup.
Step-by-step
- 1Open the Text Extractor.
- 2Upload a JPG, PNG or PDF page.
- 3Wait 5–20 seconds while the model recognises characters.
- 4Copy the extracted text or download as .txt.
- 5Fix any obvious mistakes (look-alike letters like O / 0, l / 1).
Get the best accuracy
- Shoot straight-on — perspective distortion hurts OCR more than blur.
- Use even lighting — no hard shadows across the page.
- Hold the camera steady; rest elbows on the table.
- Crop tightly to the text region before uploading.
- For scanned pages, scan at 300 DPI grayscale.
Languages we handle well
English, Hindi, Marathi, Bengali, Tamil, Telugu, Kannada, Malayalam, Gujarati, Punjabi, Urdu, Arabic, Spanish, French, German, Portuguese, Italian, Japanese, Korean, Simplified and Traditional Chinese.
When OCR earns its keep
- Digitising business cards, receipts and invoices into searchable text for your CRM or accounting tool.
- Pulling quotes and references out of a printed book or PDF scan for an essay or article.
- Capturing a slide from a conference photo and reusing the text in your own notes.
- Extracting a serial number, model number or error code from a photo of a device label.
- Indexing a stack of paper records for legal, medical or archival use so they're full-text searchable.
How to take a photo that OCRs cleanly
OCR accuracy is mostly determined at the camera, not at the algorithm. Shoot in even, diffuse light — avoid hard shadows and direct lamp glare on glossy paper. Hold the camera parallel to the page so text isn't keystoned. Fill the frame with text rather than cropping in software later — more pixels per character means more confident recognition. For handwritten notes, print-style block letters score far higher than cursive. For multi-column layouts, photograph each column separately to avoid the OCR reading across columns.
Languages, scripts and accuracy
Modern OCR models handle Latin scripts (English, European languages) at 98%+ accuracy on clean print. Devanagari (Hindi, Marathi, Sanskrit), Tamil, Telugu, Bengali and other Indic scripts run 92-96% depending on font. CJK (Chinese, Japanese, Korean) scripts work well for printed text and stumble on handwriting. Mixed-language documents (English headings + Hindi body, for example) need explicit language hints; our tool detects the dominant script automatically and you can override it if recognition seems wrong.
Improving OCR accuracy before you run it
Three quick pre-processing steps lift OCR accuracy dramatically: increase contrast so the text-to-background difference is sharper, straighten any skew so lines are horizontal (most engines lose accuracy past about 5° of rotation), and crop tight around the text block to remove background noise. Phone-scanner apps do all three automatically — using one before OCR usually beats running OCR directly on a raw photo.
FAQ
- Is the OCR tool free?
- Yes — extract text for free, no signup.
- Does it handle handwriting?
- Yes for neat block letters. Cursive accuracy varies; expect to clean up the result.
- Can I extract text from a PDF?
- Yes — upload PDF pages directly, or use our PDF to Text tool for multi-page extraction.
- Is there a file size limit?
- Around 20 MB per image. For larger scans, split into pages first.
- Are my files stored?
- No — images are processed and immediately discarded.
- What format is the output?
- Plain Unicode text, ready to paste anywhere.