Optical Character Recognition (OCR) is extensively used in automated digitisation of libraries. This task can raise several difficulties. However, since we have strong a priori on the data to be recovered (the text) we expect to be able to recover it fairly well.

In this project, we are interested in developping preprocessings in order to facilitate the OCR. Particularly, we will focus on the distortions that appear in the central part of a book (see the picture).

Things that can be done:

• Estimate and correct geometric distortions
• Compensate for the contrast
• Deblur parts that are not in-focus

The quality of the preprocessing will be quantified thanks to the success ratio of an OCR engine: Tesseract.

This work could be implemented as a java-ImageJ plugin.