A New Pattern for Handwritten Persian/Arabic Digit Recognition

Abstract

Abstract — The main problem for recognition of handwritten Persian digits using Neural Network is to extract an appropriate feature vector from image matrix. In this research an asymmetrical segmentation pattern is proposed to obtain the feature vector. This pattern can be adjusted as an optimum model thanks to its one degree of freedom as a control point. Since any chosen algorithm depends on digit identity, a Neural Network is used to prevail over this dependence. Inputs of this Network are the moment of inertia and the center of gravity which do not depend on digit identity. Recognizing the digit is carried out using another Neural Network. Simulation results indicate the high recognition rate of 97.6% for new introduced pattern in comparison to the previous models for recognition of digits. Keywords — Pattern recognition – Persian digits – Neural Network I. I NTRODUCTION ne of the major problems in handwritten digit recognition is the difference between the shapes of each digit in different handwritings. Therefore, the most active current research is using intelligent algorithms such as neural network. For recognition of digits by Neural Network, a suitable feature vector shall be extracted via the image matrix of the digits. A feature vector should have the following conditions: 1.Feature vector of two different digits should contain major differences between them as the network could conclude these differences. 2.Feature vector of a specified digit written by different handwritings should be the same, as the network could not find a meaningful difference among them. There are many different methods to extract feature vector. The coefficients of discrete wavelet transform have been used as a feature vector by Mowlaei and et al. [1]. Shirali and et al. has extracted a suitable feature vector by using shadow coding method and 32 different parts [2]. Another method for extracting feature vector has been proposed by Hosseini and Bouzerdoum by scanning the image matrix via 11 horizontal and vertical lines [3]. In this research, the idea of segmentation pattern for feature vector extraction from the image matrix has been investigated. Then for improving the feature vector characteristics, an asymmetrical segmentation pattern is introduced. Due to one degree of freedom (the control point) in the introduced asymmetrical pattern, an optimum point can be found by defining a cost function. Based on this procedure, a method of finding the optimum control point is given. Since this algorithm depends on the digit identity, for enhancing the classification results, a neural network optimizer is proposed. In this paper, all the above introduced procedures have been unified and implemented by the neural network. Part of the simulation results and the objective measure are given in this short presentation. The simulation results show very good performance and improvement with respect to other reported works.