Study Of Face Recognition Methods Based On Resampling Technology

As one of the most successful applications of image analysis and understanding, face recognition has recently received more and more attention. Problem of machine recognition of human faces continues to attract researchers from disciplines such as image processing, pattern recognition, neural networks, computer vision, and computer graphics. Face recognition system mainly include face detection, feature extraction and face recognition.In this paper, we study face recognition based on resampling technique, which consist of face recognition with semi-supervised learning method and multi-classifier ensemble. In semi-supervised learning, aiming at consistency method and its variant and combining face recognition, we study performance of face recognition with different arguments, which include factorσin exponent measures, and weight in weighting s-order measure, k in k neighbor measure and alpha in tanh measure. Moreover, we analyze the recognition's accuracy with different value in order to determine the better argument value for face recognition. In multi-classifier ensemble, we mainly study face recognition with heterogeneous classifier and homogenesis classifier. By using resampling technique to generate training set, different base classifiers (mainly include LDA, LDA plus PCA and DLDA) are obtained. In the process of resample, we take random sampling technique in each class in order to have different class's members in train set. For improving face recognition's performance, we use different fused methods, for example, voting method and sum rule etc., to fuse different classifiers recognition's results. In addition, aiming at different fused methods, results with base classifiers have two different outputs, namely class label and real value.