In this proposal, we would like to construct an automatic handwritten character recognition system based on neural network principles. A preliminary study has resulted in a model system that could recognize handwritten English alphabet and Arabic numerals in a tenth of a second on a 386 micro-computer with average accuracy over 95% in an off-line writer independent mode. The generalization is so good that it can recognize even Leviers with zigzag broad strokes that are widely distorted from any of the letter patterns that it had seen before. The system is also very simple that it takes only about 100 Kbyte of memory storage. We are planning to connect it to a text to speech system to form a totally automatic system that could read written texts out aloud. Anticipated benefits/potential commercial applications - automatic handwritten character recognition system based on neural network principles is simple, efficient, and accurate. Its performance seems to surpass greatly any commercially available system known. It can be used as a front end for a document processing system it to classify or retrieve text data automatically. With modifications, it could also be fitted into an automatic target recognition system. The commercial and military potential of this system is enormous.Key words - neural network, handwritten, recognition, character
