From scan to editable text: how OCR works and how to get good results
OCR turns a photo or scan of a page into text you can copy and edit. How it works, why it sometimes fails, and what you can do to get clean text first time.
What OCR does
A scan or a photo of a page is just an image: to a computer, black and white dots. Optical character recognition (OCR) finds the lines of text in that image, separates them into words and characters, and asks a trained model which letter each shape most likely is. The result is plain text you can search, copy and edit.
Not every PDF needs OCR. A PDF exported from a word processor or a browser already contains the real text, and a good tool reads it directly — faster and without errors. Our PDF to Word tool does exactly that, and only falls back to OCR for pages that turn out to be scans.
Why OCR makes mistakes
- Low resolution — letters that are only a few pixels high cannot be told apart.
- Blur and movement — a shaky phone photo smears the edges the model relies on.
- Skew and perspective — a page photographed at an angle has lines that are not straight.
- Shadows and poor contrast — grey text on grey paper, or a shadow across the page.
- Unusual fonts and handwriting — the further from ordinary print, the more guessing.
- Columns and tables — the text may be read correctly but in the wrong order.
How to get a good scan
- Scan at 300 dpi; for very small print, 400. More rarely helps.
- With a phone: hold it parallel to the page, use even light without shadows, and use the document mode of your camera or a scanner app.
- Crop to the page and straighten it before reading.
- Black and white or greyscale is fine; colour adds size, not accuracy.
A clean 300 dpi scan of an ordinary printed letter is usually read almost perfectly. A hurried photo of a crumpled receipt in a dim room is not, whatever software you use.
Choosing the right tool
- A photo or screenshot of text — Image to text reads it and gives you the text to copy.
- A scanned PDF that should become an editable document — PDF to Word reads digital pages directly and scanned pages with OCR.
- Tables in a PDF — PDF to Excel puts the data in rows and columns you can calculate with.
- Several photos of pages that should become one document — collect them into one file with Image to PDF first.
Our built-in OCR reads ten languages: English, Dutch, German, French, Spanish, Italian, Polish, Portuguese, Turkish and Arabic. You choose the language of the text; with Arabic, check the numbers. Digital PDFs keep whatever language they are in, because their real text is used directly.
Tables, forms and handwriting
Running text is the easy case. Three kinds of pages need extra care:
- Tables — OCR reads the words but not always the grid. Columns can be read across instead of down, and an empty cell can make the rest of a row slide one column to the left. Check the structure, not only the spelling; for tables in a PDF, PDF to Excel keeps rows and columns where it can.
- Forms — printed labels are read well, but handwritten entries in boxes are hard, and tick boxes are not text at all. Expect to type the filled-in values yourself.
- Handwriting — general OCR is built for print. Neat block capitals may work; joined-up handwriting mostly will not. Recognising handwriting is a separate, much harder problem.
A practical example: a stack of receipts
Say you want the amounts from twenty paper receipts in a spreadsheet. Photograph each one flat on a dark table in good light, from straight above, filling the frame. Read them with Image to text one by one and copy the totals; thermal receipts fade, so do it before the print disappears. For an archive, put the photos together in one PDF with Image to PDF — the images themselves remain the proof, the extracted text is the convenience.
Always proofread
OCR is never perfect, so check the result — especially numbers. Typical confusions are 0 and O, 1 and l and I, "rn" and "m", "cl" and "d". In amounts, invoices and measurements, one wrong digit matters more than ten wrong letters; compare totals with the original. A spell checker catches many of the letter mistakes in one pass.
Why it matters where the OCR runs
The documents people scan are often the most private ones: contracts, payslips, medical letters, identity papers. On Convertburda, the OCR runs in your own browser. The first time, your browser downloads the reading engine (a few megabytes) and keeps it; after that, your scans never leave your device. More about that in our guide Converting files privately.