Optimizing Recognition System of Persian handwritten digits

Features selection is one of the issues addressed in pattern recognition. By informed choice of effective features on increasing recognition rate among the overall features extracted from, the computational costs can be reduced and deducing unnecessary features are avoided. In this article, the Binary Particle Swarm Optimization (BPSO) algorithm and Binary Genetic Algorithm (BGA), both are population based algorithm series are used to find the best groups of fuzzy recognitions features of handwritten digits. Also In this article, real version of PSO (RPSO) has been used in different way to improve recognition rate. In this method, instead of choosing some of the features, one random weight has been assigned to each feature. Indeed, feature vector has been multiplied in weight vector to obtain a new feature vector. This Weight vector is obtained with RPSO. After several iterations, RPSO algorithm determines Weight set of features so that Classification accuracy increases. Fitness function in these algorithms is the number of fuzzy classifier errors and the aim is to make this value minimum. The obtained results confirmed that the population based algorithm with reducing the number of features and increasing the Rate of recognition have proper performance.