| Face recognition is a key technology in biometrics. It has become a very important research focus in the field of pattern recognition, image processing and computer vision. Algorithms with high recognition rate have important theoretical research sense and commercial applications value. With the development of information technology, Face recognition technology have play an important role in the field of identification recognition, public security, human-computer interaction, smart space and video monitoring.In this paper, face recognition technology based on contourlet transform was studied. Multi-scale and multi-direction information can be get from contourlet transform. Research shows that low frequency coefficients can describe the basic characteristics of human face very well. Meanwhile, high frequency coefficient reflects the image contour and direction information. The illumination variation of the image is not very sensitive for the high frequency coefficient, and it can be used as identification features. So both the low and high frequency coefficient are selected for face features in this paper. First, low frequency coefficient and some high frequency coefficient are selected as facial feature. These features are trained and tested by support vector machine. Second, low frequency coefficient and all high frequency coefficients are selected as facial feature. The principal component analysis is used to reduce the dimension of the facial feature. The processed feature can be used for face recognition. Experimental results show that the new methods can improve the recognition rate.On the basis of the above methods, face recognition algorithm based on nonsubsampled contourlet transform was proposed in this paper. NSCT not only have the characteristics of contourlet transform, but also have shift-invariance. Without the shift-invariance property, it can causes pseudo-Gibbs phenomena around singularities. So the directional coefficient got from NSCT can be more accurate with shift-invariance property. Because the NSCT is non-sampling, the dimension after the NSCT is more than the original dimension. In order to reduce the dimension, wavelet fusion is used to fuse the decomposition coefficients. The fused coefficients are trained and tested by support vector machine. The experimental results are very good. Also the fused coefficients can be processed by principal component analysis.In summary, after the study of face recognition method based on contourlet transform, face recognition method based on nonsubsampled contourlet transform was proposed in this paper. The new methods were experimented on the ORL face database and YALE face database, and the results are very good. |
