The OCR process – Clemens Neudecker & Niall Anderson
IMPACT work with ABBYY
IMPACT has been working with both ABBYY and IBM on OCR and Clemens starts with outlining all the steps required during the OCRing process and the extension of the FineReader Engine so it performs better with historical text.
FineReader 10 now has many improvements and modifications that are based on input from the IMPACT project.
Clemens explained the advantages of adaptive binarisation and then segmentation as they have been developed for use within the ABBYY FineReader Engine.
Adaptive binarisation can now automatically apply binarisation at different thresholds across a page to improve results.
Improved segmentation tools can now help by breaking blocks down into regions and then back down into words and then into glyphs more accurately.
Examples were shown that shows this segmentation tool in action and how the new FRengine10 has improved results over the pre-IMPACT FR engine.
These improvements are also reflected in better recognition results in FR10 than were available in earlier versions of FR.
Work is being completed on the interface so that external dictionaries can be integrated into FineReader Engine. The final aim will be for FREngine to be able to work with any language from any time-period.
Work is also going on with developing a new ALTO export format and this has been supported since FRE10R2.
Clemens’s presentation can be viewed here:
IBM CONCERT tool
Niall then gave some background to the IBM adaptive OCR process and CONCERT.
First, he provided a quick overview of some of the other ‘collaborative correction’ processes which have worked with user-feedback to improve OCR results such as the Australian Newspaper project.
Niall then demonstrated the CONCERT tools with a live online demo of the tool and by showing the screencast here:
Niall then discussed some other alternatives to OCR and post-correction using techniques such as Word Spotting.
Niall’s slides are available here: