Ocropy OCR – good results for Gothic/Fraktur Typeface

Ocropy (also called ocropus) is not a standalone program for text recognition: it consists of several command line modules for

  • binarization (creates a binary raster graphic)
  • segmentation (splitting documents into lines)
  • optical character recognition (OCR)
  • controlling the recognised text and
  • training new character sets.

Ocropy is free, modules for new fonts are relatively easy to create, and – with a little preprocessing of the images – it has very good recognition rates – up to 98% (see fig. 1).

Error rates in the detection of gothic typeface from Zedler's Unversal-Lexicon
Figure 1: Error rates of modules with different training steps for known text (training) and unknown text (testing).


Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.