Face Recognition Based On The Wavelet Transform And Singular Value Decomposition Method

The face recognition is a status identify technology which processing the person face picture and extract the effective information by using the computer, in recent years, has received the wide spread value in the pattern recognition and computer vision domain, becomes an extremely active research direction. Face recognition has the wide spread application in the identification authentication, the bank and customs' supervisory system, the public security records management and so on.This article first introduce the person face recognition research's significance, the main method and the present situation, then introduce the main face detection method and analysis the colored picture person face detection. Analysis and compare the skin color's collect character in various color space, then establish the skin color model in the YCbCr color space, in view of the definition threshold value to determine the skin color model exist the inaccuracy in the skin detecting, use the non-linear color transformation method to establish the skin color model to solve this problem, program with VC++6.0 to realize this method based on skin color face detection. The result shows that this method has better examination effect in the colored picture which has complex background.Next, introduce and analysis some person face recognition technology, include the wavelet transform, Singular Value Decomposition and K-L transformation.After that, apply the wavelet transform with Singular Value Decomposition union to the person face recognition. First deal with the person face picture with the wavelet transform, then extract character from the low frequency part by using the method which based on the Singular Value Decomposition, again classify the face picture according to the Minimum range classifier. Program with VC++6.0 to realize this method, and the result has proved that this method is feasible.Finally, using Singular Value Decomposition to obtain Eigenface character vector, and making the improvement to the theory of Eigenface method, introducing a method based on the part Eigenface. The part Eigenface method process left side and up side face respectively with Eigenface method, which based on face's symmetry and up side face's characteristic stability. Ultimately, add the two part space distance, take the face image with the smallest distance as the result. Program with VC++6.0 to realize the Eigenface method and the part Eigenface method, and compare the part Eigenface method with the Eigenface method through experiment in ORL standard face database and Yale face picture database. The result shows that the part Eigenface method is feasible, and has higher recognition rate than the Eigenface method.