From pixels to characters
A camera records pixels, not words. OCR detects regions that look like text, groups marks into lines and characters, and predicts the letters or numbers they represent.
The result can support search, copying, TXT export, document classification, or a searchable PDF layer. The original image remains valuable because it is the evidence used to check recognition.
Why OCR makes mistakes
Recognition is probabilistic. A blurred 8 may look like a 3; a fold can cross a letter; a decorative font may not resemble the forms the recognizer expects. Handwriting adds far more variation than printed type.
- Blur, low resolution, and camera shake
- Glare, shadows, stains, and textured paper
- Curved book pages and distorted perspective
- Unusual fonts, handwriting, tables, and mixed columns
- Language or script not configured as expected
OCR versus document scanning
Scanning captures and cleans the page image. OCR interprets text inside that image. A scanner can work without OCR, and OCR can be run on an existing image, but combining them creates a more useful document workflow.
Search within an app may use stored OCR data, while a searchable PDF embeds a text layer for use in other readers. These are related but distinct outcomes.
On-device versus cloud OCR
On-device OCR processes the page on the phone, which can reduce routine data exposure and work without network latency. Cloud OCR may offer different models or layout handling but requires transmitting page data to a service.
Ask an app where recognition happens, whether the original page is uploaded, how long data is retained, and what changes when an optional AI feature is used.
