Extract text from images (JPG, PNG, screenshot) using Tesseract.js OCR in the browser.
Drop in a JPG, PNG, or screenshot and this tool returns the printed text it contains, ready to copy, search, or edit. Recognition runs locally in the browser using Tesseract.js, a JavaScript port of the Tesseract OCR engine, so the image never leaves your machine. The surprise most users hit: OCR is an estimate, not a copy. Clean, upright, high-contrast printed text comes back near-verbatim; blurry photos, stylized fonts, and handwriting return errors or nothing. For a first use in a non-English language, expect a pause while that language's recognition data loads.
A concrete input and expected output from the current implementation.
Input
A screenshot cropped to a single black-on-white receipt section containing the lines: "Subtotal $12.50", "Tax $1.00", "Total $13.50".
Expected output
Subtotal $12.50 Tax $1.00 Total $13.50
Crisp, upright, high-contrast printed text is the best case for the recognition model, so each short line comes back essentially verbatim, one line per image line. Blur, skew, glare, or noise would instead produce character-level errors, which is why this example shows an idealized input.
The engine does not match whole images against templates. It first normalizes the image: grayscale conversion, thresholding to separate dark glyphs from background, and layout analysis to find lines and blocks of text. Each line then goes to a recognition model trained on printed text, which scores the most likely characters and words. Clean, flat, upright input wins; glare, shadows, and perspective all degrade segmentation and, with it, accuracy.
English recognition works out of the box. Other languages need their language data, which loads the first time that language is used, so that first run is noticeably slower than later ones. Recognition is computed locally, and processing time grows with image size and resolution. Cropping the image to just the text you need is the most effective way to speed up recognition and improve accuracy at the same time.
No. This tool is for printed text. Cursive or messy handwriting produces empty or garbled output, and even neat block-letter handwriting is recognized unreliably. This site has a separate handwriting recognition tool, built for photos and scans of written pages, which is the right choice for that input.
No. Recognition runs locally in your browser and the image is not uploaded, so it is safe to use with receipts or documents you would rather not share. The one thing that can delay a first run is loading recognition data for a language other than English, which happens once and takes a few seconds.
The example and behavioral notes were checked against the browser implementation. Standards and primary references below define the relevant format, formula, or platform behavior.
Convert between bytes, KB, MB, GB, and TB with precise binary and decimal calculations.
Open, view, and edit CSV files in a spreadsheet-style table with sorting and filtering.
Visualize JSON data as a collapsible tree with syntax highlighting and search.
Generate Code 128, EAN, UPC, and other barcode formats with customizable size and labels.
Convert images to Base64 data URLs and decode Base64 strings back to images.
Generate Apache .htaccess redirect rules for 301, 302, and rewrite configurations.