Offline Jawi handwritten recognizer using hybrid artificial neural networks and dynamic programming

This paper describes an offline Jawi handwritten recognizer using hybrid Artificial Neural Networks (ANN) as the character recognizer and Viterbi Dynamic Programming as verifier. We use a recognition-based segmentation approach to solve character segmentation problems. Segmented sub words images are segmented into a fixed width slices. The combinations of the slices form a segmentation graph. Two-layers of Back Propagation Neural Networks compute probabilities for each character hypotheses in the segmentation graph. Viterbi Dynamic Programming selects the maximum average probability of a character hypothesis combination from all possibility in segmentation graph. This system evaluates against selected word from a Jawi handwritten manuscripts. Recognition performance of the character in words presented.