Handwritten Numeral Recognition Design And Implementation Based On Touching Screen

Handwritten character recognition as an important application of pattern recognition, in recent years, is very active. And it impacts on the theory of computer, voice recognition, fingerprint recognition, artificial intelligence and so on. As the rapid development of information technology, in recent society, optical character recognition as an important method of information entry and conversion, makes the research of the character recognition having great significance. As a character recognition, handwritten numeral recognition has also been wide applications (such as bank notes, data reports, sorting of letters etc). Therefore, the success of recognition and putting into use, will make great economic and social development.Character recognition mainly uses the following ways:method of template matching, vein of statistical decision, way of syntactic structure and direct logic of law. But no matter what identification method we use, in accordance with the different characteristics can be roughly divided into two categories:first, statistical features, point density, moment, character areas, etc (mainly the overall features); second, structure features, circles, endpoints, intersection, stroke, contour, etc (mainly the detail features). In order to better solve the recognition accuracy, in the recognition process, the researchers usually combine the statistical and structure features. Because of the special question, the methods of traditional character recognition have not solving this problem.Considered that artificial neural network is a very parallel, nonlinear and high redundancy system, this article use BP network to recognize handwritten digits. The network's special structure makes that the expression, storage and process of information become very different from traditional methods. But due to the defects of traditional BP network, this paper has made many improvements, making it more suitable for dealing with the features made by elastic grid. The features involve comprehensive statistical and structural features, though, it may increase the amount of data, the extraction method is simple and the quality of features is improved. At the same time, because of the adaptive ability and anti-interference performance of neural network, it becomes probable to processing the synthesized features, to some extent, improves the efficiency and recognition rate of network.This paper introduces the creation, learning and simulation of network, based on MATLAB7.0, we use the synthesized features to train the BP network, the training and learning rate are greatly improved, then use the network to identify the samples that collected. the test means that the recognition rate has improved. Then compiling the corresponding C program, completing the design and implementation on the hardware such as the touch screen and S3C2410 processor.