An Implementation of OCR System Based on Skeleton Matching

This report gives a general review of the development of Optical Character Recognition (OCR), about its key problems and various of the techniques used. An implementation based on skeleton matching is introduced in detail, which is used mainly for printed character recognition and is insensitive to font style and size. This report emphasises on the implementation principles for text sectioning, preclassifier design, thinning algorithm, template database and matching strategy. An experiment is described to observe the system performance. Compared with conventional methods used, this implementation has included some new techniques, such as character dimension dependent broad classification method, AVP value used in fine classification and spelling check based post-processing method, in order to obtain a good performance.