Document Image Analysis (DIA) is a cross-domain of computer vision and pattern recognition and refers to an established research field dealing with the extraction of any kind of exploitable information from document images. Printed and handwritten text recognition, known as OCR/ICR (Optical/Intelligent Character recognition), is part of the discipline, but represents only one aspect. Other challenging topics include document classification, layout analysis, writer identification/authentication, signature recognition, table recognition, logical structure recognition, etc.

The aim of the Master course is to provide an overview of methods, from basic image processing to machine learning, which are described in the scientific literature to address different steps of DIA; this includes image binarization, page segmentation, graphics/text separation, text bock and text line detection, feature extraction and classification (at various levels). As a practical exercise, students will be asked to do a project (either individually or within a group of max. 4 peoples), which addresses a specific DIA challenge, including potentially the participation to international competitions.

 

- get a good overview of the DIA research domain
- get a deep understanding of the processing chains involved in DIA applications
- apply a rigorous methodology to design, implement, and evaluate a scientific experiment