Research On Ensemble Classification Based On Nearest Neighbor Multiple Classifier Selection And Application On License Plate Recognition

Although the ensemble classification of multi-classifier proved that it can improve the accuracy of recognition classification, and researchers made much achievement, how to construct an adaptive method to select classifiers, and how to make full use of the advantage of local characteristic of classifier to improve the performance of ensemble classification are still the open problems to be solved.According to above problems, based on full understanding and analysis of current research status on classifier ensemble method, we proposed a new classifier ensemble method based on classifiers selection on nearest neighbors of testing samples, and applied it to the license plate recognition. This method uses the diversity of individual classifier’ classification performance for individual sample. For individual test sample, the method uses the neighbor principle to select the classifier dynamically, and use the selected classifiers for ensemble classification. By this way, the method can adjust the selection of individual classifier dynamically in ensemble classification, and make full use of local classification advantage for individual test sample. As the experiment showed, the proposed method is effective, and the ensemble classification method showed a stable performance.According to the research content above, the main work of this paper was organized as followed:1. Introduced the research background, purpose, significance and research status at home and abroad of multi-classifier ensemble classification, the concept of ensemble classifier, classifier based on support vector machine, K nearest neighbor and its application on ensemble classification and several classic ensemble classification methods.2. Proposed a new ensemble classification method based on nearest neighbor multi-classifier selection. For individual test sample, the method selects multi classifiers dynamically, and uses the selected multi classifiers on ensemble classification. The method was proved effectively in several open datasets such as madelon, svmguide1, w6 a, ijcnn1.3. Based on our proposed method, we realized the number plate recognition algorithm based on nearest neighbor multi-classifier selection ensemble method, and we designed and realized a car license recognition system and container number recognition system based this algorithm, the experiment showed that it can improve the character recognition effectively, about an increase of 1 percent.