Why OCR Gets Text Wrong (and 8 Ways to Fix It)
By Tooldesk team · · 4 min read
OCR usually gets text wrong for one of five reasons: the text is too small, the image is blurry, the page is tilted, the contrast is poor, or the wrong language is selected. Fix the cause and accuracy jumps. Our Image to Text tool already corrects the common ones automatically, so start there, then use the fixes below for images that are still hard to read.
How OCR reads an image
OCR software doesn’t “see” text like you do. It first decides which pixels are ink and which are paper, then splits the ink into lines, words and letters, then matches each letter shape against what it has learned. Every step can go wrong:
- If ink and paper are close in colour, letters break apart or merge.
- If the page is tilted, lines run into each other.
- If letters are only a few pixels tall, their shapes are ambiguous.
- If the language is wrong, the matcher picks letters from the wrong alphabet.
8 fixes, from most to least important
1. Choose the right language
This is the first thing to check when the output is complete nonsense. Urdu, Hindi or Arabic read as English produces random Latin letters. In our tool, set Language of the text, and tick Text also contains English for mixed text.
2. Make the text bigger
OCR works best when capital letters are at least 20–30 pixels tall. A small screenshot or a photo taken from far away falls short. Our tool enlarges small images automatically before reading. If your text is a tiny part of a big photo, crop the image to the text first so the enlargement goes where it’s needed.
3. Use a sharp, original image
Every time an image is re-saved, resized or sent through a messaging app, it loses detail. Use the original photo or a fresh screenshot. On a phone, tap to focus on the text before you take the picture.
4. Straighten the page
Even a slight tilt makes OCR misread lines and break tables. Our tool detects small tilts and straightens the page, and it detects pages that are sideways or upside down. For a strongly rotated photo, you can also fix it first with Rotate Image.
5. Fix the lighting
Shadows across a page, glare from a flash and dim light all reduce contrast. Photograph documents in even daylight, without flash, holding the phone parallel to the page. Our tool stretches contrast and handles light text on dark backgrounds, such as dark-mode screenshots, by inverting them.
6. Separate text from busy backgrounds
Text printed over a photo, like a poster or a social media graphic, is one of the hardest cases. Our tool tries an extra pass that keeps only white or black letters and drops the colourful background, which recovers most of this text. Cropping tightly to the words helps too.
7. Watch out for tables
OCR engines read left to right, so a table can come out as one long line or with rows missing. Our tool finds tables with visible lines and reads each cell separately, then puts tabs between columns. Tables without lines are read as text with spaces.
8. Scan at the right resolution
For paper documents, 300 DPI is the standard for OCR. Go up to 400 DPI for very small print. Lower than 200 DPI loses letter detail, and far higher just makes files bigger.
Quick troubleshooting table
| What you see | Likely cause | Fix |
|---|---|---|
| Random symbols and Latin letters | Wrong language | Pick the right language |
| Most words right, some letters swapped | Slight blur or small text | Use the original image, crop to the text |
| Lines merged or out of order | Tilted page or columns | Straighten, or crop one column at a time |
| Nothing found | Very faint text or text in a logo | Increase contrast, crop, or type it by hand |
| Table rows missing | Borderless table | Crop the table and check each row |
0 read as O, 1 as l |
Ambiguous shapes | Proofread numbers carefully |
When OCR isn’t the answer
Some text can’t be read reliably: decorative fonts, handwriting, text curved around a logo, and characters only a few pixels tall in a low-resolution image. If accuracy matters, for example for an invoice total or an ID number, check those parts by eye.
For the basics of copying text on each device, see our guide to copying text from an image. For whole scanned documents, read how to convert a scanned PDF to Word.
Questions people ask
What is a good OCR accuracy?
On a clean, straight, 300 DPI scan of printed text, a good OCR engine gets nearly every character right. Phone photos, small text and unusual fonts lower that, which is why the fixes above matter.
Does a higher resolution always give better OCR?
Up to a point. Going from 150 to 300 DPI helps a lot. Going beyond 400 DPI rarely helps and makes processing slower.
Why does OCR work on my phone but not online?
Phone tools like Live Text use models tuned for camera photos. If an online tool struggles, check the language setting and crop the image to the text.
Can OCR read text in any font?
Standard fonts are read well. Script, decorative and very condensed fonts are much harder.