Handwriting Digit Recognition With Neural Network Ensemble

Handwritten digit recognition is widely applied to such field, as post code, statistic table, bank note and so on. In these fields involved in account and fiance, digit recognition request a high rate of correctness, and a less rate of error, the same time, processing much of data require a quick speed of system running. Many ways which is perfect in theory, but speed is slow in processing is impractical. Therefore, it is a challenge task to research a high performance handwritten digit recognition system.Because of its good generalization capability, the research about Netural network ensemble is being a hot point.A handwritten digit recognition system based AdaBoost algorithm was introduced in this paper. AdaBoost can construct a highly accurate classifier by combining many weak classifier that just had slightly better accurate than random prediction. The paper discuss a realization how to use AdaBoost algorithm to handwritten digit recognition. The following is done in this paper.(1)Nine features for handwritten digits based on macroscopical, partial and microcosmic are extracted, which are applied in nine respective neural networks.(2)Make some improvements on BP neural network to quicken the network constringency speed and to avoid fake saturation phenomenon.(3)Make a solution for handwritten digit recognition applied AdaBoost algorithm, the results of experiment show that this system can get high correct rate at a certain extent. It is worth to be studied later.