Handwriting and OCR: what to expect, and what not to

Handwriting is where automatic reading has improved most in recent years, and also where it still makes the most mistakes. Both are true at once, and anyone who works with handwritten documents needs to know both.

What changed

Older OCR systems recognised one letter at a time, and with handwriting that barely worked: everyone writes letters differently, and the letters touch each other. Newer models read words and lines in context. When one word is unclear, the context of the sentence, or of the field on a form, helps fill it in. That is why neat handwriting reads much better today than it used to.

What reads better

• Printed forms filled in by hand, especially when each value is written in a box or on a line.
• Separate block letters, in Hebrew and in English.
• Clearly written numbers: amounts, dates, phone numbers.
• Dark ink on light paper, in a sharp photo with even light.

What is still hard

• Cursive and joined-up writing, especially very personal handwriting.
• Similar digits: 1 and 7, 4 and 9, 0 and 6. Here a small mistake changes an amount.
• Faint pencil, writing over a printed background, and crossings-out.
• Notes in the margins, and lines written at an angle or between other lines.
• Signatures: they are not read as values, and they should not be.

How to work with it

1. Expect marks. In a handwritten document, more values are marked for review. That is good: the mark tells you exactly where to look.
2. Check every value that matters. Amounts, dates, ID numbers and account numbers: against the original, digit by digit. Clicking a field marks its place on the page.
3. Correct before exporting. A value corrected in the workspace is the value that goes to Excel.
4. Improve the photo. With handwriting, the difference between an average photo and a good one is especially large. See the photo guide.

When not to rely on automatic reading

When one wrong character can cause harm, as in a legal document, a prescription or an account number for a transfer, automatic reading is a draft, not a result. Use it to save typing, and check every value against the original. And if the whole document is in very personal handwriting, typing it by hand may be faster than checking and correcting.