The Recognition Of Handwritten Numeric Characters Based On Neural Network

In the field of pattern recognition, the study show that only using a method is often difficult to identify a complex classification problem well, but different classification methods often have highly complementary nature. the integration of multiple classifiers can significantly improve the recognition rate, Therefore, the research of integrated method of multiple classifiers has become a hot spot in recent years.In this paper, an integrated neural network of Handwritten Digit Recognition system has been built, the system mainly consists of three sub-networks each sub-network can be trained Independently, after training, each sub-network can test Numeric characters, After the parallel of the three sub-network,a method named" Posterior probability based on the weight of Automatic Adjustment Act" have been used to identify the sample of Numeric characters Throughout the system, the main work of this article, as well as innovations in the following areas:1 In the macro, micro, and the transform domain aspects, three feature of Numeric characters have been extracted The first sub-network is based on the overall characteristics of the macro, and the second sub-network is based on the micro-edge features, the third sub-network is based on the Fourier transform transform domain features.2 According to the characteristics of the given characters in this experiment, after appropriate transformation, a large number of basic concepts and methods in the field of digital image processing are used in this topic. a variety methods of image preprocessing are used, such as:using the methods of digital image processing field to enhance the edge effects of the Numeric characters, de-noising, character gray-scale transformation and so on, these practices improve the character recognition rate for a certain3 In the integrated approach, this paper proposed the "posterior probability based on the weight of automatic adjustment Law" linear integration method. According to independent experiments of each sub-network, can be observed that to identify the different characters different sub-networks have different effects, according to this thinking, in the parallel integration, the sub-network which have the best effect in the recognition process will be given the greatest weight In this paper, the neural network integrated digital handwriting recognition system, have used a variety of pattern recognition methods In an integrated manner, reflects all aspects of handwritten numeric characters features fully。The test results show that the system has improved the accuracy of handwritten numeral recognition, but, there are many things worth of study...