Face Recognition Technology Research Based On Nonsubsampled Contourlet Transform

Face recognition is generally described as given a static or dynamic image, to identify one or more persons by the existing face image databases. Face recognition has become a classic and hot topic in the field of pattern recognition and artificial intelligence research currently. Exploring the high recognition rate of face recognition has important theoretical significance and application value.Wavelet transform has good time domain and frequency domain characteristics. It can express and analyze the image signal very well and be recognized as a new technology in the field of information acquisition and processing internationally. Wavelet transform has excellent features in one-dimensional signal processing. However, this advantage can not be simply extended to two-dimensional space or higher dimensional space. Wavelet transform in the lack of tansform direction makes it can not fully describe the image characteristics and will miss large information in the transformation. Contourlet transform is a new multi-scale geometric analysis method, which not only has the multi-resolution characteristics of wavelet transform and time-frequency localization properties, but also has a strong orientation and anisotropy.Nonsubsampled Contourlet Transform is improved Contourlet Transform to solve the spectral leakage and aliasing defects due to the subsampled Contourlet Transform can not meet the translation invariance. It has a better performance on the details of the image features in the image.In this thesis, a method based on Nonsubsampled Contourlet Transform and Kernel Fisher Discriminant Analysis(KFDA) for face recognition is proposed. In the method, the features of the image in Nonsubsampled Contourlet Transform combined with KFDA are extracted and classificated for face recognition. The difference of recognition rate and recognition time are researched for face recognition between Nonsubsampled Contourlet Transform and Contourlet Transform. At the same time, the other method that based on contourlet transform and Support Vector Machine (SVM) for face recognition is proposed. The experimental result shows that combining the Nonsubsampled Contourlet Transform with KFDA which overcome the shortcomings of Contourlet transform better can get the details of the performance characteristics of the image better and obtain excellent recognition rate. Because of the SVM needs to solve problems of the secondary regulation, the stage of training sample has higher time complexity than KFDA approach. In the large training set conditions, KFDA calculates eigenvectors of the matrix only and has small amount of computation, therefore the method has obvious advantages such as less time consuming, more efficient feature extraction and higher classification accuracy.