Farsi Handwritten Digit Recognition using Multi Expert Classifiers

Abstract

Ensemble multiple weak classifiers is a new method in order to gain higher recognition rate in pattern recognition problems. In this paper we propose a new efficient method for recognizing Persian handwritten characters. This method uses three different algorithms to extract forty features from a template as a feature vector. Every feature vectors which are extracted from any of those algorithms, are transferred to an independent neural network. The outputs of the neural networks transfer to one other neural network to decide for final result andits task is mapping the input to the correct class. Experimental evaluation on a real working environment database with 9000 samples demonstrates theconfidence to our proposed multimodal scheme and its accuracy against unimodal methods. The recognition rate is 96.4% in the best situation that progress the previous methods about 1.55% in recognition problems.