PDF OCR — Extract Text from Scanned PDF

OCR (Optical Character Recognition) turns images of text inside a PDF into actual selectable, searchable text. Scanned documents, camera photos of receipts, and image-heavy legacy PDFs all benefit — after OCR, you can copy text, search within the file, and pass it through other tools like PDF-to-Word. The hardest part of OCR is the language coverage: recognizing English is very different from recognizing Chinese, Japanese, or Arabic script. This tool supports over 100 languages via Tesseract.js, though larger language packs (like CJK) download additional model data the first time they are used. OCR runs entirely in your browser. The Tesseract WASM engine processes each page, adds a text layer to the PDF while keeping the original image visible, and delivers a searchable output. Nothing leaves your device.

Last reviewed: July 2026

What is this?

OCR happens in three stages. First, image preprocessing normalizes the page — binarization converts color to black and white, deskewing corrects tilted scans, and noise removal cleans up specks. Second, layout analysis segments the page into text regions and non-text regions (images, lines, tables), then within text regions identifies individual lines and words. Third, character recognition uses an LSTM (Long Short-Term Memory) neural network to convert word images into character sequences. Modern OCR is remarkably accurate on clean, high-DPI scans of printed text — often above 99% for English. Handwriting, low-resolution images, and complex layouts drop accuracy significantly.

How to use this tool

1. 1Choose a PDF containing scanned or image-based pages. If your PDF already has selectable text, OCR is not needed — use the Extract Text tool instead to skip processing time.
2. 2Select the language(s) in the PDF. Choose only the languages actually present — running OCR with unused languages slows processing without improving accuracy. For mixed-language documents, select the primary language.
3. 3Click Process. The tool loads the language model (first run may take 10-30 seconds to download the ~10MB model), then processes each page. Expect 2-5 seconds per page on typical scans.
4. 4Download the OCRed PDF. The output looks identical to the input, but you can now select, copy, and search text throughout. Try Ctrl+F to verify text is recognized correctly.

Tips & best practices

• OCR accuracy depends heavily on scan quality. 300 DPI scans of printed text produce 98-99% accuracy. 150 DPI drops to 85-90%. Camera photos vary widely — 90-95% for good lighting, 70-80% for poor.
• The first run downloads the Tesseract WASM engine (~15MB) plus language models (5-15MB each). Once downloaded, models are cached — subsequent runs are much faster.
• For CJK languages (Chinese, Japanese, Korean), the language pack is significantly larger (~15-25MB per language). The first run takes longer than for Latin-script languages.
• Handwriting recognition is not what this tool is optimized for. Tesseract handles printed text well but struggles with cursive script. Accuracy drops to 30-60% on handwriting.
• OCR adds a hidden text layer under the visible image. If the output PDF looks unchanged, that is expected — try selecting text to confirm OCR worked.