An Efficient Feature Extraction Method for the Middle-Age Character Recognition

In this paper, we introduce an efficient feature extraction method for character recognition. The EA strategy is used to maximize the Fisher linear discriminant function (FLD) over a high order Pseudo-Zernike moment. The argument, which maximizes the FLD criteria, is selected as the proposed weight function. To evaluate the performance of the proposed feature, experimental studies are carried out on the historic Middle-Age Persian characters. The numerical results show 96.8% recognition rate on the selected database with the weighted Pseudo-Zernike feature (with order 10) and 65, 111,16 neurons for the input, hidden, and output layers while this amount for the original Pseudo-Zernike is 93%.