Current Optical Character Recognition (OCR) technology is considerably better than that of five years ago, but it is still not at a level of accuracy needed for widespread practical applications. If OCR systems are to be a useful alternative to skilled human typists, the error rates of the machine must be reduced by one to two orders of magnitude. This project applies new approaches to the problem. The problem is here approached by modeling the various steps in the character recognition process and then developing analytical techniques for each step. The project will continue research on extracting features directly from gray scale images (as compared with the usual method involving binarization of the images before recognition). The research will also address construction of character prototypes and development of flexible graph matching algorithms taking into account both typeface specifications and the characteristics of distortions and noise encountered in the printing process.
