Language-Based Feature Extraction Using Template-Matching in Farsi/Arabic Handwritten Numeral Recognition

A recognition system based on template matching for identifying handwritten Farsi/Arabic numerals has been developed in this paper. Template matching is a fundamental method of detecting the presence of objects and identifying them in an image. In the proposed method, templates have been chosen so that they represent the features of FARSI/Arabic prescribed form of writing as possible. Experimental results show that the performance of proposed language-based method has been achieved more than the other usual common feature extraction approaches. NM-MLP is used as a classifier and trained with 6000 samples. Test set includes 4000 samples. The recognition rate of 97.65% was obtained, which is 0.64% more than Zernike moment approach.