Method For Face Recognition Using Curvelet Transform

For the multiresolution ideas of wavelet transform are quite useful for analyzing the information content of facial images, wavelet transform is wild used in face recognition. Recently, some new multiresolution tools such as contourlet transform and curvelet transform are raised. However, little attention has been paid to them. Curvelet transform is a new kind of multiscale geometric analysis method, which not only has the multiresolution decomposition and time-frequency properties, but also has strong directional elements and anisotropy properties. Owning to these properties, curvelet transform could capture the geometric features of images effectively.In the paper, the concept of image multiscale geometric analyze and the basic theory of curvelet transform are introduced. And the distribution features of curvelet coefficients and energy statistics are analyzed. Then the following works have been done.Firstly, the available face recognition methods are introduced. To analyze the same face database, the principe components analysis (PCA) and the linear discriminant analysis (LDA) theory are used. Through simulation experiment, the advantage and weakness are compared between these algorithms.Secondly, a face recognition algorithm based on curvelet transform is proposed. The algorithm includes the following three aspects:(1) Pre-process a face image and extract the curvelet transform coefficients; (2) Use PCA and LDA to reduce the dimension of the face image; (3) Classify the coefficients after dimension reduction coefficient, then recognize the face image. In order to study the effects of expression and light to the face recognition, the ORL face database and Yale face database are used, and the recognizing results of the algorithm are compared with that of the recognition algorithm based on wavelet transform. The experimental results show that the low-frequency coefficients generated by curvelet transform can well reflect the invariability of face pose and the facial feature, and the rencognizing rate of the algorithm is better than that of the algorithm wavelet transform.