Troubleshooting
What Is OCR and How Does It Work?
Optical character recognition explained without the jargon — the stages of the process, what makes it fail, and what it is genuinely good for.
OCR — optical character recognition — turns a picture of text into text. It is the technology that lets you search a scanned contract, and the reason a photographed receipt can be filed as data rather than as a snapshot.
What happens, in stages
- 1Pre-processing: the image is straightened, contrast is normalised and noise is removed.
- 2Layout analysis: the page is divided into text blocks, columns, images and tables.
- 3Line and word segmentation: each block is split into lines, then into words and characters.
- 4Recognition: each character shape is classified, historically by feature matching and now by neural networks.
- 5Post-processing: dictionaries and language models correct improbable results — "rnodern" becomes "modern".
The last stage does a lot of quiet work. Raw character recognition is noticeably worse than the final output; language context is what makes it usable.
What makes it fail
| Cause | Why it hurts |
|---|---|
| Low resolution | Character shapes merge and cannot be separated |
| Skew | Line segmentation misreads where lines begin and end |
| Faint print | Thin strokes vanish, changing the shape |
| Background patterns | Watermarks and shading confuse the character boundaries |
| Unusual fonts | Decorative and script faces are far from the training data |
| Handwriting | A different problem entirely, and a much harder one |
| Wrong language setting | The language model corrects towards the wrong words |
How it is used in a PDF
The recognised text is added as an invisible layer over the original scan rather than replacing it. You keep the exact appearance of the document and gain search, selection and accessibility. If recognition made a mistake, the visible page is still correct — which is a considerably safer design than replacing the image with recognised text.
Where it stands today
Modern neural OCR is very good on clean printed text — accuracy in the high nineties per cent — and handles multiple columns, tables and mixed languages far better than the systems of a decade ago. Handwriting recognition has improved but remains unreliable for anything you cannot check.
It has not become a solved problem. It has become good enough that the remaining errors are concentrated exactly where they are hardest to spot: names, numbers and codes.
Do it now
OCR PDF
Extract text from scanned documents. It runs in this browser tab — your file is not uploaded anywhere.
Open OCR PDFFrequently asked questions
Is OCR accurate enough to trust?+
For search and reading, yes. For figures and reference numbers, verify against the image.
Does OCR work on photos?+
Yes, though a photograph of a page is harder than a scan. Use your phone's document scanner mode for a much better result.
What resolution is best?+
300 DPI. It is the balance point between accuracy and file size.
Can OCR read tables?+
It reads the text. Reconstructing the grid is layout analysis, which is separate and less reliable.