Line Handwritten Chinese Character Recognition BP Neural Network And HMM Hybrid Model Is Based

With the continuous development of the society, we often need to deal with a lot of text,report forms and so on in our Daily life. Using manual keyboard input is not able to fully meet the information needs of people, how to use the computer to identify text quickly and accurately has become increasingly important. As a branch of character recognition (OCR) technology, off-line handwritten Chinese character recognition technology has become popular in the pattern recognition.Artificial neural network has a lot of advantages, such as good fault tolerance, strong adaptation and the ability to imitate human intelligence. So it can handle the situation of complicated environment, unclear background and undefined inference rules. Therefore applying the neural network theory to handwritten Chinese character recognition has great theoretical significance and practical value. In this article, the neural network basic principle, basic method and relevant algorithm are studied. Then aiming at the BP neural network algorithm’s slow learning algorithm convergence and local minimum value problem, we use the additional momentum method and make better the error function to improve better the BP algorithm. While BP neural network is not good at modeling, we design and implement an off-line handwritten Chinese characters recognition system based on BP neural network and HMM aiming at its disadvantages. BPNN-HMM hybrid model can not only make full use of the BPNN’s good classification ability, but also make up for the inadequacy of BPNN modeling ability by taking advantage of the HMM’s powerful time-domain modeling capabilities.Therefore this hybrid model has strong modeling capabilities, classification ability and adaptive ability.In this article, the steps and the corresponding methods and the algorithm of the preprocessing part was introduced. Then we introduced median filter for de-noising and threshold binarization algorithm in detail. In Characteristics processing parts, we introduce two feature extraction methods, the structure feature extraction and the statistical feature extraction. In this paper,we implement the feature extraction method based on the projection boundary chain. On this basis, we proposes a handwritten recognition method which is based on BP neural network and hidden Markov model to improve the recognition performance.At last, the simulation experiment was carried out in the MATLAB(R2009b) environment by recognizing10Chinese characters. The experiment results show that using the method this paper proposed to do off-line handwritten character recognition is feasible, and the recognition accuracy rate is up to89.0%, which is better than some existing methods.