Face Recognition Based On Contourlet Transform And PCA

Face recognition techonology is a pattern recognition techniques by processing face images and extracting the feature of face to idengtify the person using compter.It's a human identification techonology based on the inherent organism's characteristics of human bodys.Because it is safer,more effective and reliable,more and more people pay attention to it. Face recognition consists of three parts:preprocessing, face representation and classification.Face representation is extracting the feature of face,it's the most critical part of face recognition. In this thesis, we use Contourlet transform to reduce the image dimensions in face image preprocessing, and then use principal component analysis for feature extraction and finally recognize the face by distance classification.Principal component analysis is a feature extraction method based on mathematical statistical, it is widely used in face recognition and other field of pattern recognition. However,in one hand a single PCA face recognition method is influenced by the changes in light conditions and facial expression,so the recognition rate is affected.In the other hand,in the PCA face recognition method,a face image may be considered as a vector of dimension N 2,the dimension of the vector is high,so the covariance matrix of the face samples is great,and the computation of feature extraction is huge.For the lack of PCA face recognition method,in this thesis,we proposed a method of face recognition based on Contourlet transform and PCA. Contourlet transform is a new image multiscale geometric analysis method, it can capture the image texture and contour information of image better than wavelet transform.In this thesis,we use Contourlet transform to obtain the low frequency sub-band of face image and to remove high frequency nosie such as facial expression and so on before the PCA method,and then we use PCA method to the low frequency sub-band of face image.In the image preprocessing by Contourlet transform,it can not only reduce the dimension of images,but also can extract more efficient features for recognition,so the recognition rate is improved.In the finally,we do a lot of experiment on the orl faces and the yale faces,and analyze many kinds of parameters in the experiment,and get the influnces of recognition by the parameters.The result of the experiment shows that the recognition in the method proposed by this thesis is higher than that in the tranditional PCA face recognition method.