Research On Algorithms Of Face Segmentation And Recognition BAsed On Wavelet Transform

Face recognition is a kind of biometric identification technology possessing great development potential, researching on the technology of face recognition has great theoretical and practical values. In recent years, the technology of face recognition based on two-dimensional wavelet transform has achieved unprecedented progress. This dissertation researches the theory and practical technology of face recognition based on two-dimensional wavelet transform thoroughly.The preprocessing of the face image must be first researched before face recognition. The preprocessing of the face image is to change the face images collected by the equipments into the standard images. The process contains three steps:face detection and eye location, geometry normalization, grey value normalization. This dissertation emphasizes on the research of face detection method which is based on Gadabouts statistical learning.Two-dimensional wavelet transform and its response characteristic in recognition applications are researched. Two-dimensional wavelet transform is realized by computing the convolutions of a bank of two-dimensional filter and the grey values of pixels in an area around a given position in an image. In this dissertation, the result of the research has proved that the part characteristic of the face can been represented by choosing the parameters of wavelet filter Recognition based on two-dimensional wavelet transform is better than that based on the grey of the original image directly.Classical elastic bunch graph matching algorithm is improved and the elastic bunch graph matching face recognition algorithm based on template bunch is proposed. The elastic bunch graph matching algorithm uses the labeled graph which its nodes are expressed by a set of two-dimensional wavelet transform coefficients which can describe the part characteristic of the face to represent the face image. These nodes are located in the position where it is meaningful to the recognition on the face image. Every edge of the labeled graph is labeled with the metric information which can describe the relevant position of the adjacent-nodes. And the grid image structure composed with all of the edges can describe the geometry characteristics of the whole face. The computation load of classical elastic bunch graph matching algorithm is large and the recognition velocity of it is slow, in this dissertation, seven representative grid structures are obtained through clustering of the grid structures of the face labeled graph of many training images, these grid structures are used to constitute a template bunch of the face bunch graph. Combine the elastic bunch graph matching algorithm with the learning method of Gadabouts during the process of matching. First of all, locate the position of the eyes, and make geometry standardization of the input image with the eyes' coordinate being the datum mark; then, select the most suitable grid structure from the template bunch to determine the geometry characteristic of the face. Last, precise matching is performed further based on the outcome.Test proves that the elastic bunch graph matching face recognition algorithm based on template bunch not only can reduce more operation than the elastic bunch graph matching, but also can consumedly improve the velocity and accuracy of recognition. It is a doable and better face recognition method.