Using Single-Layer neural network for recognition of isolated handwritten Persian digits

Recently a new method for recognition of isolated handwritten Persian digits, based on support vector machines (SVMs), has been introduced. In this research, this method was implemented for the same task with three new modifications, i.e. only one popular shape was considered for digits written in different shapes; sizes of glyphs normalized to digit boundaries; MLP (Multi-Layer Perceptron), SVM/MLP and SLP (Single-Layer Perceptron) neural networks used for classification. Each digit is considered from 4 different views, and from each view, 50 features were extracted to obtain 200 features. Multiple SVM classifiers were trained to separate different classes of digits for MLP and were compared with ordinary training for MLP and SLP. Experiments on real life samples from AMIRKABIR database showed that the proposed features, when used with SLP, could perform very well leading to a recognition rate of 98.3%, compared with 94.14% obtained from MLP classifier with SVM training using the CENPARMI database.