On-device recognition — the OCR engine and selected language data download, but the source file is not uploaded.

Online Free OCR — Scanned PDF and Image to Text

Short answer: To extract text from a scanned PDF or image, run OCR and review the recognised text before copying or downloading it. QuickMerge runs OCR in your browser and supports seven document languages without uploading the file.

Turn scanned PDFs and images into selectable, copyable text in seven document languages. Recognition runs entirely in your browser, so nothing is uploaded and there is no account to create.

Free to use
No sign-up
7 languages
Runs in your browser
No uploads

How do I extract text from a scanned PDF or image?

Select an image or scanned PDF to read (up to 100MB):

Drag & drop an image or PDF

JPG, PNG, WebP, or a scanned PDF · one file at a time

Document language:

Read Text From Any Scan or Photo — Without Uploading It

Local recognition avoids a QuickMerge document upload. It does not establish recognition accuracy or suitability for regulated records. Review optional analytics choices, your organization's requirements and every critical value in the result.

How to Extract Text From a Scan or Image

Add your file. Drop in a photo, screenshot, or scanned PDF. Images (JPG, PNG, WebP) and multi-page PDFs are both supported.
Pick the document language. Choosing the right language dramatically improves accuracy. You can even combine two (for example English + French) for mixed documents.
Click Extract Text. The first run downloads the language model to your browser (a few seconds); after that it is cached. A progress bar shows recognition page by page.
Copy or download. The recognized text appears in the box below — copy it to the clipboard or download it as a .txt file. Your original file is never modified.

What OCR Is Good At (and Its Limits)

Use sharp, straight, high-contrast printed text in the correct language. Blur, skew, faint strokes and unusual layouts can cause recognition errors even when the image looks readable. A 200–300 DPI scan can help preserve detail, but resolution does not establish accuracy; compare names, identifiers and amounts against the source.

Common Uses

Make a scanned PDF searchable: pull the text out of a scan so you can search, quote, or reuse it.
Digitize receipts and invoices: get the numbers and vendor details into text, or use Image to Excel when the source is a structured table.
Capture text from a photo: a whiteboard, a book page, a sign, a slide — anything you photographed.
Accessibility: convert an image-only document into text a screen reader can announce. For a text-based PDF that needs editable DOCX output, use PDF to Word.

How It Works & Why It's Private

When you click Extract Text, the tool loads the Tesseract OCR engine (WebAssembly) and selected language data into your browser. For a PDF, each page is rendered to a canvas locally with pdf.js; an image is read directly. Tesseract analyzes the pixels on your device and returns plain text. The engine and language assets come from public CDNs, but the source document is not uploaded. Use the Proof of Privacy steps to inspect the network requests yourself.

Next steps

Before this step

straighten the pages first — skewed text is the most common cause of poor recognition.
crop to the text you want — so nothing else competes for attention.
turn PDF pages into images — when you want to process page by page.

What to do with the result

get a table rather than plain text — if what you scanned was tabular.
paste recognised rows into a spreadsheet — when the OCR output is already columnar text.
save the extracted text as a PDF — to keep a clean, searchable copy.

Browse all OCR and extraction tools, or see every QuickMerge tool.

No completed size benchmark is published for this tool. See measurement coverage and limitations; stated upload limits are application guards, not tested device capacity.

Frequently Asked Questions

Is my document uploaded for OCR?

Recognition runs locally through WebAssembly. The engine and selected language data must be available to the browser before recognition. Initial loading needs network access; cached offline availability is not guaranteed.

Which languages are supported?

The dropdown offers English, Spanish, French, German, Italian, Portuguese and Dutch, plus English + French and English + Spanish for mixed documents. The underlying Tesseract engine supports many more, but those are the ones this tool ships with. Choosing the correct language noticeably improves accuracy.

Can it read handwriting?

No. The engine is trained on printed text. Handwritten and cursive material will not produce usable output.

Why is the first run slower?

The first time you use a language, the browser downloads its model (a few megabytes) and caches it. Subsequent runs in the same browser are much faster because the model is already stored locally.

The result has mistakes — how do I improve it?

Use a sharper, higher-contrast, level image, and make sure the selected language matches the document. Straight 200–300 DPI scans give the best results; blurry or skewed photos give the worst.

Is there a file size or page limit?

The upload control accepts a source file up to 100MB. That is an application guard, not proof that your device can process a file of that size. Decoded pages and recognition models also use memory. Start with a small sample or process a few pages at a time; no completed size benchmark is published for this workflow.

Prepare recognition and check its output

This workflow extracts characters from page images. Recognition errors remain possible even on a clear scan.

Use a clear source

Straighten the page and crop irrelevant surroundings without cutting off text. Small characters, blur, skew and complex layouts can reduce recognition quality. Increasing resolution cannot recreate detail missing from the source.

Tables need alignment checks

For a table use Image to Excel and compare every important cell position. Empty cells and wrapped labels can shift recognized values into neighboring columns. Reconcile identifiers and totals against the image.

Keep the image beside the text

The extracted text is a draft, not proof of accuracy. Correct names, numbers and punctuation before reusing it. Saving recognized text with Text to PDF creates a new layout; it does not attach a verified text layer to the original scan.

Avoid degrading the input first

Run recognition from the clearest available source. Lossy compression or downscaling can remove character detail. A particular DPI does not guarantee successful OCR.