OCR
Converting images of text (scans, photos) into machine-readable text strings.
Think of it like
Transcribing a handwritten letter so you can search and edit it.
Example
Scan a receipt; OCR extracts "Total: $42.50, Date: 2025-01-16"; you can now search for this expense in your records.
How it actually works
Combines vision (detect text regions) and sequence modeling (recognize characters). Classic approaches (Tesseract) struggled with curved text and low quality; modern deep learning (CRAFT + attention-based sequence-to-sequence) is more robust. End-to-end models learn bounding box detection and character recognition jointly.
For product teams
Unlocks automation: invoice processing, document search, accessibility (read aloud).
For engineers
Text detection network (YOLO-style) finds text boxes; character recognition network (sequence-to-sequence) decodes characters within each box.
Related
- Subtask of.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome