Based On Local Entropy Retention And Statistical Feature Extraction Method Of Face Recognition Research

In numerous personal identification methods, the one based on facial image has become a hot and important research point because of its special superiorities and broad application prospects. Furthermore, among many facial-image-based methods, the global-statistic-based method gets much attention for its clear concept and simple computation. However, the method based on global statistics often has two disadvantages. On one hand, it treats the feature in each dimension of the training images equally. On the other hand, it does not use the category information of the training images in learning. Since these two disadvantages have affected the recognition rate of this method, to overcome them this paper first proposes a entropy-based local feature preserving method, then integrates this method with the principal components analysis method, the completely two-dimensional principal components analysis method and the kernel principal components analysis method, respectively, to form the corresponding human face recognition methods. The contents and the innovation results of this paper are summarized as follows:1. The uncertainty of each dimension in the image set of the same person is smaller than that of different persons. According to this characteristic, this paper proposes the entropy-based local feature preserving method, which first generates coefficients corresponding to the uncertainty of each dimension, then use these coefficients to perform linear projection. The proposed method can effectively reduce the redundancy and the disturbance data in human face image.2. The principal components analysis method is a linear unsupervised method based on the global statistics, it treats the feature in each dimension of the training images equally and it also discards the very useful category information of the training images in learning, thus it cannot obtain a better recognition rate, to overcome these disadvantages, this paper integrates the entropy-based local feature preserving method with the principal components analysis method, and proposes the entropy-based local preserving principal components analysis method, the experimental result indicates that the proposed method obtains a higher recognition rate than the principal components analysis method.3. The complete two-dimensional principal component analysis method is a linear unsupervised method based on the global statistics too. This paper integrates the entropy-based local feature preserving method with the complete two-dimensional principal components analysis method, and proposes the entropy-based local preserving complete two-dimensional principal components analysis method, the experimental result indicates that the proposed method obtains a higher recognition rate than the complete two-dimensional principal components analysis method.4. The kernel principal components analysis method is promotion of the principal components analysis method in the non-linear space, but it is still a unsupervised method based on the global statistics. This paper promotes entropy-based local feature preserving method into the non-linear space and proposes the entropy-based local preserving kernel principal components analysis method, the experimental result indicates that the proposed method has a remarkable promotion.