An Unconstrained Handwritten Numerals Recognition Way Based On Combined Classifier

In this paper ,a handwritten recognition system is set up . This system includes learning part and recognition part . In the learning part, three BP neural networks are trained . there are four steps in the recognition part. They are preprocessing, getting character, deleting fault character and combined recognition . During the realizing of the system , the following is done in this paper.1. In this paper, many preprocessing ways of handwritten numeral are used. For example proving, smoothing, binaryzation and thinning. In this paper, many improvements on arithmetic are presented .The results of experiment show that the improvements are valid.2. A new way to get the inflection point is brought forward.. It is to get the inflection point by using the least square imitate multinomial arithmetic .The result of experiment show that the method is right.3.in this paper , the way of the traditional deleting fault character is improved . during the realization ,we find the improved way can delete the fake port, isolated point and fake thripoint more validly .4.BP neural network arithmetic is used in the handwritten numeral recognition system. In order to reflect the whole character, we use many input modes to train the networks.5.Make some improvements on BP neural network to quicken the network constringency speed. For example , change the learning-factor dynamicly, avoid fake saturation phenomenon.6. A combined classifier is designed. There are two recognition ways used in the classifier. They are the least distance pattern recognition method and the BP neural network pattern recognition method. There are two steps in the combined classifier .The least distance pattern recognition method is used to classify roughly in the first step. Three BP networks connected paralled is used to classify finely.7.Many combined arithmetic are used to calculate the last recognition result according to the results of the three BP networks .They are the average arithmetic and the raising one's hand arithmetic. The result of the experiment show that the raising one's hand arithmetic can improve the capability of the handwritten numeral recognition system .A handwritten numeral recognition system based on the combined classifier is built up in this paper. Many pattern recognition ways and many handwritten numeral charactersare used in the system . The results of experiment show that this system can get high correct rate . It is worth to be studied later.