Research And Improvement Of Nearest Feature Classifiers

The classification process of pattern recognition requires the classifier to be usingto classify the data. The classifiers play the role of the decision-making mechanism inthe classification process. The classifiers are very critical to the final categorydiscrimination, the performance of the classifier, being pros or cons, will be directlyrelated to result of pattern recognition, being good or bad. Several pattern recognitionclassifiers including artificial neural networks, support vector machines, Adaboostclassifier and the nearest neighbor classifier.In this thesis, improve the nearest feature classifiers are the main content for theface recognition. Based on the classical nearest neighbor classifier, the nearest featureclassifiers use the geometric characteristics for extending the number of the limitedsample to achieve better recognition performance. There are three relativelyrepresentative nearest feature classifiers. They are nearest feature line (NFL) classifier,nearest feature plane (NFP) classifier and nearest feature space (NFP) classifier.However, they have the disadvantages of extended imprecisely and the highcomputational complexity. This paper proposed several improved algorithms for thesedrawbacks.The main contribution of this thesis is as follows. Firstly, based on the nearestfeature line classifier, several improved algorithm are proposed, which achieve betterresults in applications such as face recognition, palmprint recognition and objectrecognition; Secondly, based on the nearest feature plane classifier, an improvedclassifier is proposed, which achieves better results in applications such as facerecognition, palmprint recognition and object recognition.