Letter pattern recognition

A knowledge-based system that automatically recognizes the components of letters and stores an OCR (optical character recognition) version of each letter is described. The system first digitizes the document and segments it into blocks using only low-level segmentation techniques, then recognizes the block text contents and finally recognizes blocks as components. It uses attributes such as relative position, size, and contents to do so. The system has a highly efficient pattern-matching method, based on a novel block matrix representation of relative position information. The rule-based knowledge and pattern-matching functions are integrated in a C-language system. On a sample of 70 letters, the prototype system correctly recognized 89% of positively identified components