Research Of Implementation Of Support Vector Machine And Neural Network In Handwritten Numeral Recognition

OCR is one of the important branch of pattern recognition while HNR is one of the important branch of OCR. Because there are few types of numbers, so many algorithms are proposed or improved from applying to HNR. In addition, with the development of OA and finance, precise and efficiency of HNR needs to be enhanced, so the author chose So, it is of great significance for researching HNR.The mainstream algorithms to enhance the performance of the classifiers is to do some optimization to the classifiers or to search for the classifiers which has better performance, to integrate classifiers to enhance the performance has not been widely researched, so this particle try to enhance the performance of the classifiers by intergrading the classifiers.The algorithm of the particle apply Artificial Neural Network(ANN) and Support Vector Machine(SVM) to set up a two-layer handwriting numbers recognition system. When setting up the first-layer classifier, generates the intergrated classifier based on the optimized AdaBoosting algorithm; then generates the binary SVM classifiers based on the theory of binary search. The system achiedves to search for the corresponding SVM classifier to make the final classification, which is based on the result of the first-layer classifier.The main job of the particle will be illustrated in the coming paragraphs.In the topic of feature extracting, proposes the net scanning PCA algorithm to extract the principal comment of the image as the input feature of intergrated ANN classifier, and proposed the incremental kernel PCA algorithm to extract the incremental kernel principal component feature as the input feature of binary SVM classifier.In the topic of intergrated ANN classifier, chooses PNN classifier to construct the intergrated ANN; the generates the weak classifier series based on the optimized AdaBoosting algorithm; finally generates the studying rule based on the weightful voting rule.In the topic of SVM classifier, set up the binary classification decision tree based on the theory of binary classification, the extract the pattern and train the SVM classifier based on the decision tree. In the process of setting up desion tree, applys kenel clustering algorithm which based on the particle optimizing algorithm to search for the global optimal clustering solution in the Hilbert space.The particle designs and implement the two-layer HNR algorithm which is composed of intergrated ANN classfier and binary SVM classifier, then proves that the algorithm indeed enhance the performance of the HNR system.