The Research On Face Recognition Technology Based On Curvelet And PCA-class Method

After nearly fifty years development, human face recognition technology has made a great progress and has been applied in many practical fields. But when some conditions such as illumination, gestures and expressions are not ideal, recognition accuracy suffers dramatically. So how to eliminate effects of these changing conditions has became the most challenging task in the face recognition. One of the most important techniques used to solve this problem is to combine wavelet transform with PCA-class method. However wavelet transform mainly reflects the point singularity of images, it cannot accurately express the information of curve edges in face images. To overcome this weakness, curvelet transform is proposed and become more popular. Because its basic support space is anisotropic, it can better represent the fuctions of curve edges in high-dimensional spaces. This thesis tries to study the better method based on curvelet transform and PCA-class method. The mainly works in this thesis are as follows:(1) The basic knowledge of face recognition and curvelet transform is studied. The thesis introduced the background and significance of face recognition, then summarized and analyzed the development in this field at home and abroad. The wavelet transform and curvelet transform, which are both the commonly image processing technologies in face recognition area, were studied in detail. For the disadvantage of wavelet transform, the thesis proposed to apply curvelet transform to face recognition, and analyzed coefficients which got from the decomposition of the image by curvelet transform.(2) An improved (2D)2PCA method is proposed. The thesis described the methods of PCA class, included PCA,2DPCA and (2D)2PCA. For the sensitivity to illumination change and the standard measure of matrix distance, a improved (2D)2PCA method was proposed. The proposed method-firstly introduced an exponential decayed factor to decay the illumination of image samples, and then extracted the features with (2D)2PCA, finally adopted the nearest neighbor classifier based on p matrix distance to realize classification. Experiments on Yale database showed that the improved algorithm got better results.(3) A novel algorithm based on curvelet transform and improved (2D)2PCA is proposed. The newly algorithm decomposed face images to get low frequency coefficients by curvelet transform, and combined with the improved (2D)2PCA to extract face feature matrix and get recognition results. Throught the compared analysis, simulation experiment results showed that the proposed algorithm had higher recognition accuracy and shorter recognition time, and it was also robust to illumination changes.