Efficient and Intelligent Off-Line Handwritten Recognition Systems

The major contributions of the thesis are listed below: ¿ Feed forward neural architecture is identified as the most suitable classifier for recognizing handwritten English characters. ¿ A new zonal feature extraction, called diagonal feature extraction is proposed. ¿ Hybridization of feature is investigated to enhance recognition accuracy. The best hybrid feature set is identified. ¿ Novel training strategy for neural classifier is proposed to improve the average and worst case recognition accuracies. ¿ A comprehensive CRS for Mixed handwritten English alphabets is built. ¿ A general algorithm for designing CRS is developed. ¿ A methodology to validate the performance of the designed CRS is evolved and illustrated. This thesis presents a systematic procedure for designing and developing a high accuracy character recognition system for handwritten English characters.