Recognition of a Subset of Most Common Persian Words Using Zernike Moments and Back-Propagation Neural Network

Optical character recognition (OCR) is a popular research topic in artificial intelligent area. One of the most important parts of OCR is word recognition. So in this paper, we propose a combination method of selected subsets of Zernike features and MLP Back-Propagation Neural network to recognize Persian words. These words are the most useful and common words among 1000 words in Persian handwritings. We overcame sensitivity problem, scale changes and rotation of words with different handwritings in the process of recognition. We select 60 out of 91 Zernike features to get a better result besides using Feed-forward back propagation neural network classifier (BPNN). Furthermore we experiment within different value of inputs to set a proper alpha and momentum to achieve accuracy of the BPNN. In our project the recognition accuracy is between 78%-94%.