About

OCRopus(tm) is a state-of-the-art document analysis and OCR system, featuring pluggable layout analysis, pluggable character recognition, statistical natural language modeling, and multi-lingual capabilities.

The OCRopus engine is based on two research projects: a high-performance handwriting recognizer developed in the mid-90's and deployed by the US Census bureau, and novel high-performance layout analysis methods.

News

2011-10 Refactoring is nearly complete, with OCRopus now divided into a number of well-defined native code modules (ocrorast, ocrolseg, ocrofst) and high-level Python code (ocropy).

The top-level repository (ocropus) is now a repository that you can check out and that should contain everything needed for building OCRopus.

2011-05 There has been significant refactoring and cleanup over the last year.

OCRopus is now effectively usable as a NumPy library with native NumPy arrays
most of the APIs are documented through the Python interfaces
Unicode and ligature support largely works
all recognition can now be carried out from Python
there are top-level commands for recognition and training written in Python
classifiers now can cope with large character sets
there are tools for clustering and correcting character shapes
there is support for ligatures
there are numerous bug fixes
training is possible on very large datasets (many millions of samples)

Plans

What remains to be done before the next official release:

remove a lot of unused C++ code and consolidate iulib and ocropus C++ code (DONE)
factor out some C++ and Python libraries into separate projects (DONE)
retraining of all the models
the language models need to be tested and improved

Next steps:

integration of book-adaptive recognition tools into the Python code
integration of new layout analysis that works for distorted pages
integration of new orientation detection
integration of new text/image segmentation code

Resources

OCRopus Mailing List (subscribe / contribute)
OCRopus Group Pages (add your contributions here)
User-contributed links and resources (add links here)

Related Projects

iulib Library (you need to install this)
hOCR Tools -- tools for manipulating OCR output
DECAPOD -- camera-based document capture and tagged PDF generation
PyOpenFST -- Python bindings for OpenFST (for language modeling)

Documentation

The following is the most important documentation:

Release Notes -- summary information about releases
Development Install -- how to install the development version of OCRopus
Using -- some information about how to use OCRopus
Training -- how to train OCRopus
Publications -- information about algorithms

If you want to contribute to the primary documentation, please check out hg clone https://wiki.ocropus.googlecode.com/hg and submit patches against the documentation.

Additional links you may find useful are here:

C++ Programming -- extending OCRopus in C++
C++ Coding Conventions -- memory management, pointers, naming, formatting
File Formats -- file formats used by OCRopus
Book-Level Representation -- directory layout for whole book recognition
hOCR Output Format -- (X)HTML-compatible OCR output format

Bugs / Issues / Enhancements

Please use the "Issues" tab above to submit bugs, feature requests, etc.

When submitting bug reports, please keep the following in mind:

include OCRopus version/hg changeset, OS version, compiler version
sample images that fail (tag with SampleImage if you attach an image)
stack trace from GDB if you can get that

Until the beta release (version 0.5) we mainly care about "big stuff" bug reports and failing documents; minor compile issues or cross-platform issues don't matter that much right now. Please also only recognition failures on fairly clean scanned documents for the time being.

Contributing

If you want to contribute code to OCRopus, or if you have a patched version or variant, please use Google's Server Side Clone Support for Mercurial. You can maintain your own variant, add experimental features, etc., and share your patches/changes easily with others even if we haven't incorporated them into the main branch yet.

Acknowledgements

The system is combining the work of many contributors and previous projects. The core developers work at the IUPR research group at the DFKI and gratefully acknowledge funding by Google and the BMBF TextGrid project.