The creation of accessible texts which are primarily available in a non-editable format (e.g., PDF, JPG files) is the most difficult task. At this stage, Optical Character Recognition software (OCR) are very important for obtaining editable formats that can be made accessible.
Currently, there exist many OCR working on normal texts with optimal performances. On the other hand, many software and studies on the recognition of on-line handwritten formulae can be found. However, the recognition of on-line formulae is a different problem with respect to recognition of printed formulae processed in long documents. For instance, in the case of on-line formulae, additional information can be exploited, such as hand motion, information not available in printed documents. Presently, the only software able to process whole documents with printed characters and recognize both text and formulae is InftyReader, developed by Japanese universities and designed for necessity of visually impaired people. However, since there exists a single OCR of this kind, research in this field is still poor and not enough exploited. Thus, research in this field could open to new horizons and new frontiers.
An OCR software has two main components: a pattern recognition algorithm and an image segmentation algorithm.
The project aims at developing original algorithms of this kind finalized to realization of an OCR that automatic recognizes both text and formulae. We are developing the pattern recognition algorithm by using artificial neural networks. The image segmentation algorithm will be developed using some classical methods completed by a fuzzy approach and studying techniques based on the calculus of variations.