How to copy text from an image with OCR, and get better results
Lecture slides as screenshots, a photo of a paper notice, an announcement posted only as an image: sometimes you need the text, not the picture. Optical character recognition (OCR) turns the text in a photo into text you can copy. Accuracy depends heavily on the photo, so here are tips to get cleaner results.
Extracting text from an image
- Add a photo or screenshot to the Image to Text tool below. Pasting with Ctrl+V works too.
- Choose the recognition language. Select more than one if the text is mixed.
- Start recognition. You will see progress, and the first run takes a little longer while language data downloads.
- Review the result, then copy it or download it as a TXT file.
Tips for better accuracy
Shoot straight on, in good light
Photos taken at an angle distort the letters. Hold the camera parallel to the page and avoid shadows.
Crop to the text you need
Pictures, patterns and other paragraphs in the frame add stray characters. Crop with the Image Cropper so only the text remains.
Turn on grayscale and contrast
For yellowed paper or faint text, the preprocessing options separate letters from the background more clearly.
Make small text bigger
If the text is only a few pixels tall, recognition struggles. Moving closer so the text fills more of the frame helps most.
Where OCR struggles
| Case | Why |
|---|---|
| Handwriting | Everyone writes differently, so accuracy is far lower than for printed text |
| Decorative fonts | Unusual strokes get read as other letters |
| Tables and columns | Cells may come out in a mixed order |
| Text over photos | Background patterns can be read as letters |
Always read the result once and compare important details such as numbers, names and amounts with the original. Look-alike characters such as 0 and O, or 1 and l, are common mix-ups.
Safe for private images
Recognition runs inside your browser, so your photo is never uploaded. You can use it on receipts or ID copies. Just think about where you paste the resulting text.