Recognition of Farsi Handwritten Cheque Values Using Neural Networks

The subject of word recognition has been receiving considerable attention in recent years due to the increasing dependence on computer data processing. Several methods for recognizing Latin, Chinese words have been proposed. However, works on recognition of Farsi words has been relatively sparse. Techniques developed for recognizing other language can not been used for recognizing Farsi words. In this paper, we introduce a new approach for recognition Farsi handwritten cheque value based on sub-word segmentation and neural classification. In our method, at first we did some preprocessing algorithms such as binarization and one pixel noise removal and then segment each word of cheque into its sub-word at second step. After segmentation, structural features such as loops, dots are extracted. Our feature vector consists of 13 elements, which is as input of neural network. In classification, a MLP network with Back Propagation algorithm is used. For increasing results of our approach, we suggested a post-processing step which is based on grammar of cheque sentences. Our results show that we can recognize 58% of sub-words in classification step, and 85% in post processing. Our results are related to worst case forms and with better one, we can get better results