Home Page

• documentation, source code
  
• project pages
• list of contributors
• list of resources

OCRopus License

• very widely used license
• similar to BSD license
  
• far more liberal than GPL/LGPL
• use it for open source project
• use it for closed source products
• if you distribute source, you need to acknowledge its source
• if you assert patents against other users, you can't use it anymore
• see the license for details
  

Target Applications

• document indexing and search
• on-line reading
• re-typesetting
• editing
• information extraction

State of OCR

• OCR is lots of different things...
• physical layout analysis
• logical layout analysis
• text recognition
• font recognition, style recognition
  
• math recognition
• table recognition
• all of these problems are unsolved for unrestricted input
• what works fairly well...
• physical layout analysis for simple layouts
• text recognition for alphabetic languages and clean inputs
• OCR is harder than speech recognition
• expectation of 100% accuracy
• human-designed medium, rather than natural medium

State of Commercial OCR

OCR Tradeoffs

• throughput and memory requirements
• average error rate
• frequency of catastrophic failures
• types of errors
• layout analysis errors
• OCR errors correctable by language models
• OCR errors on names / identifiers
• cost

OCRopus Architecture

OCRopus Goals

• performance
• significantly reduce average character error rate
• greatly reduce undetected catastrophic failures
• meet production throughput/memory requirements
• functionality
• any script, any language
• pluggable architecture
• testable architecture
• fully statistical foundation