Research On Printed Character Recognition System Based On BP Neutral Network

This article is mainly concerned about the auto-recognition of printed English letters and digits in image.In the system's design, this paper introduces the neural network pattern recognition technology and presents a design work of Character Recognition System based on BP neural network. In this paper, the whole system is divided into three main modules (pretreatment, feature extraction and crude classification, and the character classifier based on BP) for a detailed description.To improve the performance of the system and to reduce the error rate and rejection rate as much as possible, this article carefully analyses the key and difficult problems which were encountered in the design of these three main modules, and proposes the following solutions:1. In the design of Pretreatment module, a variety of image processing technology is concerned. A series of algorithm including the removal of discrete miscellaneous noises, character segmentation and slope adjustment is proposed. The realization of these algorithms lays a solid foundation for Character Feature Extraction.2. Character Feature Extraction is the key point, which decides the success of the overall design. After comparing several popular feature extraction methods, a rough classification method based on the features of closed curves and vertical lines in character skeletons is presented. It evenly divides the original aggregate including 62 characters into three subsets to lower the difficulty of follow-up treatments. This paper brings about an algorithm of extraction which combines rough grid feature and regulated projection feature. It balances overall and partial features of characters and makes identity of similar English letters easier.3. In the design process of Characters Classifier based on BP, we carefully studied the key problems in network design which consists of network framework design, parameters design, network training and network recognition, and put forward a set of network design options to optimize network performance. The effectiveness of the optimization program was verified in the final system test.This paper shows that character recognition system based on BP neural network is not only able to effectively recognize characters in the same font of training samples, but can also recognize characters in other fonts. In addition, it has certain capacity of anti-interference and anti-deformation. This design program is not only a reference to improve the performance of OCR systems, but can also be used in other fields related with OCR.