OCR a PDF, extract text from any scan in your browser
Turn scanned or image-only PDFs into editable text. Seven languages, page-by-page confidence, copy or download as .txt. No upload, no sign-up, no watermark. First run downloads a one-time language model.
- English, Urdu, Arabic, Hindi and 3 more
- Per-page confidence score
- Download as .txt
- Works offline after first load
100% secure. Your PDF is never uploaded. First run downloads a language model (one-time).
- 100% Secure
- No Upload
- 7 Languages
- Page Confidence
- Works Offline
- No Sign-up
Languages supported
Pick the main language in your document before running OCR. Using the wrong model is the biggest cause of bad results.
| Language | Script | Direction | Notes |
|---|---|---|---|
| English | Latin | Left to right | Best accuracy of all languages in the model set. |
| Urdu | Arabic (Nastaliq) | Right to left | Handles printed Nastaliq well; handwriting is hit-or-miss. |
| Arabic | Arabic | Right to left | Modern Standard Arabic and printed dialects. |
| Hindi | Devanagari | Left to right | Printed Devanagari; excellent for newspapers and textbooks. |
| Spanish | Latin | Left to right | Handles accented characters (á, é, í, ó, ú, ñ, ü). |
| French | Latin | Left to right | Handles accents (é, è, ê, à, ç, ô, î) and cedillas. |
| German | Latin | Left to right | Handles umlauts (ä, ö, ü) and the sharp s (ß). |
How the OCR works under the hood
A quick look at what the tool actually does with your PDF.
| What | Detail |
|---|---|
| OCR engine | Tesseract.js (WebAssembly build of Tesseract OCR) |
| Where OCR runs | Entirely in a Web Worker inside your browser |
| Page render scale for OCR | 2.0x (approximately 144 DPI) |
| Language models | Downloaded once per language, then cached by the browser |
| First-run time | Usually 10 to 20 seconds for the model download |
| Output format | Plain text with "--- Page N ---" separators |
| What is not supported | Handwriting, mathematical notation, complex tables, floor plans |
How OCR actually works
Short answers to the questions that decide whether your OCR result will be good or bad.
| Question | Answer |
|---|---|
| What OCR does | Looks at each pixel region, matches patterns against a trained model, and guesses the character. |
| Why resolution matters | OCR needs enough pixels per character. Screen-resolution PDFs (72 DPI) are too small; 144 to 300 DPI works well. |
| Why skew hurts accuracy | Rotated text is hard to match against the model. If a scan is tilted, straighten it before OCR. |
| Why confidence matters | Low-confidence pages usually mean blur, noise, or unusual fonts. Treat those results with caution. |
| Why language matters | Using the wrong language model produces gibberish, even on clean scans. Pick the actual language of the document. |
| What searchable PDF means | A PDF with an invisible text layer behind the scanned image. This tool outputs plain text, not a searchable PDF. |
Why this is the best PDF OCR tool
Most online OCR tools upload your scanned PDF, cap you at a handful of pages per day, and put a watermark on the result. This one does not.
| Feature | TToolDeck | Typical online OCR | Desktop OCR apps |
|---|---|---|---|
| No sign-up or email | Yes | Partial | No |
| No upload, PDF stays on your device | Yes | No | No |
| No page or file size cap | Yes | Partial | No |
| No daily OCR quota | Yes | No | No |
| Multiple languages in one tool | Yes | Partial | Yes |
| Right-to-left scripts (Urdu, Arabic) | Yes | Partial | Partial |
| Page-by-page confidence score | Yes | No | Partial |
| Download as plain .txt | Yes | Partial | Yes |
| Copy to clipboard in one tap | Yes | Partial | Yes |
| No watermark on output | Yes | No | Partial |
| Works offline after first load | Yes | No | No |
Comparison reflects publicly available information at the time of writing. Features may change. Representative examples of common patterns.
More free PDF tools
Free browser-based PDF tools. Nothing uploaded, ever.
PDF to Word
Convert PDF into editable .docx
Merge PDF
Combine multiple PDFs into one
Split PDF
Split by page, range or every N pages
Compress PDF
Shrink PDF size for email and uploads
Extract Images from PDF
Pull every embedded image out of a PDF
Sign PDF
Draw or type a signature and place it on any page
PDF to JPG
Render every page as a high-quality image
Image to PDF
Turn JPG and PNG into a single PDF
OCR PDF
You are hereRead scanned PDFs and get editable text
How it works
- 1
Drop your PDF
Drag and drop or click to choose a scanned or image-only PDF. The file is opened straight from memory, nothing is uploaded.
- 2
Choose language and run
Pick the main language in the document, then click Run OCR. The first run for a language downloads a small model (one-time, cached after).
- 3
Copy or download the text
Review the page-by-page results with confidence scores, then copy to clipboard or download as a .txt file named after your PDF.
Everything you get for free
Runs entirely in your browser
OCR is done locally using Tesseract.js compiled to WebAssembly. Your scanned PDF, the images inside it, and the recognized text never leave your device. Nothing is uploaded to a server.
Seven languages
English, Urdu, Arabic, Hindi, Spanish, French and German, including right-to-left scripts like Urdu and Arabic that many OCR tools handle badly. Pick the main language in the document before running.
High-resolution page rendering
Pages are rendered at 2x scale (roughly 144 DPI) before OCR, which is the sweet spot for accuracy without blowing up memory. Small print and dense text come out much cleaner than at screen resolution.
Per-page confidence score
Every page shows its own recognition confidence percentage. If a page is blurry, skewed or photographed at an angle, the score drops and you know exactly which pages to re-scan.
Page markers in output
The extracted text is separated with clean "--- Page N ---" markers, so you can drop the .txt into an editor and immediately know where each page ends. No re-parsing needed.
Copy or download
One-tap copy to clipboard, or download the whole result as a .txt file named after your PDF. Use whichever fits your workflow: pasting into Word, or saving for later.
Handles encrypted PDFs
PDFs that open without a password are processed automatically. If the PDF is locked, run it through the Unlock PDF tool first, then bring the unlocked file here to OCR.
No watermark on output
The extracted text is the raw OCR output, no branding, no "made with ToolDeck" line, no forced header. Copy it into a document and it is ready to use.
What people use it for
Digitize scanned contracts
Turn signed contracts, leases and agreements that only exist as scans into editable text. Copy into a template, a CRM, or a searchable archive.
Extract text from old books
Public-domain books, historical documents and out-of-print material often exist only as scanned PDFs. OCR them to get searchable, quotable text.
Read receipts and invoices
Paper receipts and printed invoices photographed into PDFs become text you can sum, sort or paste into accounting software.
Urdu, Arabic and Hindi documents
Legal, religious and government documents in Urdu, Arabic or Hindi are still often scans. OCR them with the correct language model to get editable text.
Searchable research notes
Academic PDFs that were scanned from print cannot be searched or copied. OCR them once and keep the .txt alongside the original. Future-you will thank you.
Language learning material
Extract text from language textbooks, worksheets or exam papers in Spanish, French or German. Use the .txt as flashcards, translation input, or study notes.
Questions people ask
What does OCR actually do?
OCR stands for Optical Character Recognition. It looks at the pixels of a scanned page or image, finds patterns that look like text, and converts them into actual editable characters. Without OCR, a scanned PDF is just a picture of text, you cannot search, copy or edit it. With OCR, you get a .txt file you can paste into Word, a spreadsheet, or a database.
Is my scanned PDF uploaded to a server?
No. The PDF is opened with pdf.js in your browser, each page is rendered to a canvas, and OCR is run using Tesseract.js (a WebAssembly build of the Tesseract engine), all on your device. Nothing is uploaded, nothing is stored, nothing is logged. The only network request is the one-time download of the language model on first use.
Which languages does the tool support?
Seven: English, Urdu, Arabic, Hindi, Spanish, French and German. This covers the most common printed-script needs, including the right-to-left scripts (Urdu, Arabic) that many OCR tools mishandle. Pick the main language in your document before running. Using the wrong model is the single biggest cause of bad OCR results.
Why is the first run slow?
The first time you run OCR in a given language, the browser has to download the trained model for that language (usually 5 to 20 MB). This is a one-time cost, after that the model is cached and later runs start instantly. You will see "Loading OCR engine (first run may take 10-20s)..." during this step.
How accurate is the OCR?
For clean, straight, high-resolution printed text, accuracy is typically 95 to 99%. Accuracy drops sharply with: low-resolution scans (below about 150 DPI), skew or rotation, handwriting, low contrast, background patterns, unusual fonts, or mixed-language documents. Every page shows its own confidence score so you know which pages to trust.
What does the confidence score mean?
Tesseract returns a per-page confidence percentage between 0 and 100, based on how certain it is about the characters it recognised. Above 90% is usually clean text. 70 to 90% means mostly right, spot-check it. Below 70% means expect errors: re-scan at a higher DPI, straighten the page, or improve contrast.
Does it work on handwriting?
Not well. Tesseract is trained primarily on printed type. Neat, well-separated handwriting sometimes comes through, but cursive and messy handwriting are usually beyond it. For handwritten documents you need a specialised service like Google Cloud Vision, AWS Textract or Azure Computer Vision, none of which run fully in a browser.
Can it make a searchable PDF?
Not directly. This tool outputs plain text (.txt). If you specifically need a searchable PDF (a PDF with an invisible text layer behind the scanned image so PDF readers can search it), you will need a dedicated tool. The text you extract here is exactly what goes into that invisible layer, so you can also paste it into a tool that builds searchable PDFs.
Does it work on password-protected PDFs?
PDFs that are encrypted but open without asking for a password are handled automatically. If the PDF requires a password to open, unlock it first with a PDF unlocker, then bring the unlocked file here to OCR.
What about tables and columns?
Tesseract reads text in a left-to-right, top-to-bottom order within each page. Multi-column layouts, tables and sidebars often come out interleaved. If your document is table-heavy, this is expected. You will need to reconstruct the structure yourself. OCR gives you the words; layout has to be fixed by hand or with a table-aware tool.
Does it work on my phone?
Yes, the tool opens on any modern iPhone or Android browser. But OCR is CPU-heavy, and phones take longer per page than a laptop. For a 2-page document you will barely notice; for a 50-page scan, a desktop is much faster.
Is there a page or file size limit?
No hard limit in the tool. The practical limit is your device memory and patience. A 200 MB scan with 500 pages will work on a modern laptop but may be slow. There is no daily quota, no sign-up, no watermark, and no cost.
Your scanned PDF never leaves your device
The PDF is opened, rendered and OCR'd entirely in your browser using a local copy of the pdf.js engine and the Tesseract.js OCR engine compiled to WebAssembly. Recognised text is held in memory only until you copy or download it, then freed. The only network request is the one-time language model download on first use. Read how we keep it private.