A variation and distortion tolerant structural pre-classifier for hierarchical character recognition

This paper presents a structural pre-classification approach to reduce the number of class models to be compared with a given sample during the main classification stage. This helps to increase the overall classification speed and to improve the recognition accuracy. The approach is based on preparing high-level coarse shape models of character classes permitting similar character-classes to merge into super-groups. The approach is robust to distortion and font or writing style variations.