A New Approach to Feature Selection in Handwritten Farsi/Arabic Character Recognition

Feature extraction and feature selection are very important steps in pattern recognition systems. However, finding an optimal, effective, and robust feature set is usually a difficult task. In this paper, with the use of a comprehensive study on offline handwritten Farsi/Arabic digit recognition systems, a set of well-known features were extracted. Then, by employing one- and two-dimensional spectrum diagrams for standard deviation and minimum to maximum distributions, an optimal subset of initial features set was selected automatically. Experimental results, according to one of the biggest standard handwritten Farsi digit datasets, the HODA, had shown 95.70% accuracy with the proposed method. The achieved results showed a salient improvement in system precision in comparison to using other state-of-the-art approaches.