Character Recognition Algorithm Based on Neural Network

Abstract

Character recognition is a topic with great research value and it includes all typical issues in the field of pattern recognition: data acquisition, processing and selection, selection of input sample expression as well as the guided training on the recognizer with sample set. Artificial neural network (ANN) is a mathematical model which simulates the behavior characteristics of neural network and performs distributed and parallel information processing. Along with the progress in neural network and logic techniques, people have used many new methods and means on the study of character recognition. This paper proposes a character recognition method based on improved BP neural network. This method extracts features of training samples according to certain methods, builds the classification analysis system, divides each character to be recognized into 8*8 blocks for digitalization and represents each block with a vector. The input vector is defined as an input vector matrix and the vector represents a certain character or number. Then, it conducts feature extraction on certain unknown sample to be detected with the same method and the system will judge its classification according to the sample feature extracted. The training of neural network shall train the arrays which represent characters or numbers respectively in the input end under supervision and correspond to the specific location in the output end. The test result on this character recognition method shows that this algorithm has relatively high recognition accuracy, a faster processing speed and strong robustness in the applications.