Batch OCR & Search

• Can an existing piece of text be mapped to the layout from an original image?
  • Tesseract can support this.
  • Very valuable for old manuscripts

• OCR error rate depends on quality of the writing and quality of a scan.

• Some people are using Hadoop to distribute some of the OCR and analysis.

• Possibility of author tracking (via handwriting style tracking) being integrated into mainstream tools?

• Possibility of revision tracking built into the OCR’d contents metadata?

• How much OCR metadata to include?
• Human input vs automation vs hybrid approach.
• Linking OCR’d content back to catalog records is straightforward and easy approach.

• hOCR - Standard output format for OCR’d content. http://en.wikipedia.org/wiki/HOCR

• OMR (optical music recognition) - http://en.wikipedia.org/wiki/Music_OCR

Projects mentioned:

• ActiveOCR Corrections are fed back into the software and it learns to become more accurate.
• Tesseract
  • Lots of languages
  • Skew on images can produce poor results
• OCRopus