Optical character recognition (OCR) can save hours of retyping, but results vary wildly. A crisp scan may come out almost perfect, while a dim phone photo returns nonsense. The difference is rarely the software. It is the input. These nine tips explain what the engine needs.
1. Start with enough resolution
OCR engines recognise shapes of letters, so letters need enough pixels. For scans, 300 DPI is the usual recommendation. For screenshots, characters should be roughly 20–30 pixels tall; if they are smaller, enlarge the image before recognition or use an upscale option.
2. Use contrast, not colour
Dark text on a plain, light background is ideal. Colour backgrounds, gradients and textures confuse the step that separates text from background. Converting to grayscale and boosting contrast often helps with faded print or uneven lighting.
3. Fix the lighting and shadows
When photographing a page, use even light, avoid casting your own shadow and keep the phone parallel to the paper. Shadows across a page can make half the text unreadable even though it looks fine to you.
4. Straighten and crop
A tilted page, curved book spine or visible table edge adds noise. Crop to the text area and rotate until lines are level. Even a few degrees of skew reduces accuracy.
5. Choose the right language
The language model supplies the vocabulary and character set the engine uses to decide between lookalike shapes. English selected for a Malay or French page produces avoidable errors. For mixed content, choose two languages if the tool allows it.
6. Pick the right layout mode
Automatic layout analysis works for ordinary pages. For a single paragraph, a block mode avoids the engine seeing columns that are not there. For a single line, such as a number plate or caption, a line mode is much more reliable. For signs, posters and labels with scattered words, sparse text mode finds text in odd positions.
7. Avoid lossy compression
Heavily compressed JPEGs add blotchy artefacts around letters. If you can, screenshot as PNG or scan at high quality. Re-saving an image repeatedly makes it worse.
8. Know the hard cases
Handwriting, stylised fonts, tiny print, text over photos and complex tables are difficult for general-purpose OCR. Do not expect perfect output; plan for a proofreading pass instead.
9. Proofread what matters
OCR errors are systematic. Watch for 0 versus O, 1 versus l versus I, 5 versus S, rn versus m, and missed punctuation. Numbers, names, dates and amounts deserve a second look. After extraction, use the Text Formatting Cleaner to rejoin broken lines and its find-and-replace option to fix a repeated misreading in one go.
Privacy considerations
Many online OCR services upload your image to a server. Fakulti’s Image to Text runs the recognition inside your browser, so documents with personal or financial details stay on your device. Only the OCR engine and language files are downloaded, and your browser caches them after the first use.
Try it: Image to Text (OCR)
Extract text from photos, screenshots and scans in your browser.
Open the tool