Optical character recognitionOCR

Turning the text in a photo of an ID into data fields.

A person photographs their ID, and the software reads the words and numbers in the image and fills in fields such as name, date of birth, and document number.

Pattern-matching each character in a document image against known glyph shapes to reconstruct machine-readable text, then structuring the result into named fields like name or date of birth. A glyph is the visual shape of a letter or digit. Modern systems use machine learning models trained on many fonts and capture conditions.

An OCR pipeline typically follows these steps:

1. The system finds the document in the image and corrects tilt and perspective.
2. It locates regions of text.
3. It recognizes each character.
4. It assembles characters into words and assigns them to fields.

The reading is harder than it looks because of common problems:

• Glare and shadows that hide characters.
• Blur and low resolution that blend neighboring characters.
• Decorative backgrounds and security patterns that confuse the reader.
• Similar-looking characters, such as the letter O and the digit 0.

Optical character recognition sits next to terms that are easy to mix up:

• Data extraction is the broader step of turning a document into structured, scored fields, and OCR is its core technology.
• Machine-readable zone is a standardized block of text that OCR reads on passports and ID cards.
• VLM (Vision-Language Model) is a newer kind of AI model that reads a document's layout and meaning, not just its characters.

For more, see The History of Optical Character Recognition (OCR) and What is OCR for Identity Verification?