Online Recognition Of Handwritten Persian Words Using A Novel Hierarchical Fuzzy System

Persian (Farsi) handwriting is inherently cursive and variable in style. Moreover, many Persian characters have similar body but different secondary strokes e.g. dots. As a result, Persian handwriting recognition is an extremely complex task. Many researchers have tackled the problem by different approaches; however, the research in this field is still in its infancy. In this paper, we propose a novel hierarchical system for recognition of Persian words. As a preliminary step, words are divided into subwords based on pen-up and pen-down information. Then, a three level fuzzy recognition method is applied to recognize each subword and each word is recognized by merging the recognition results of its subwords. In the highest level of subword recognition, subwords are divided into characters using a backtracking method with pruning. In the mid level, each character is recognized using a dynamic programming approach. In the lowest level, the segments of each character are classified. The result of each level is used in its above level. Fuzzy if-then rules and fuzzy inference is used in all three levels. Given the segments of each word, fuzzy feature vectors of the segments and the fuzzy-rule base for character recognition, the proposed method is hierarchically optimal, i.e. the system finds a word with maximum membership with respect to the hierarchy we used. Our preliminary experiments with the proposed system show satisfactory results.