Efficient retrieval of face images based on curvelets and singular value decomposition

Face Recognition Technology is one of the fastest growing field in the biometric industries. Although human seem to recognize faces with relative ease, machine recognition of faces is a challenging task. In this direction the paper proposed here is an automatic face recognition technique based on curvelets and Singular Value decomposition (SVD). Curve discontinuities present in the face images are very well captured by curvelet transform coefficients with different scales and orientations. Here 4 scale and 8 orientations of curvelets have been used. The extracted features still form a high dimensional subspace. To further reduce computational complexity SVD on curvelet transform coefficient is applied to get the optimized feature vector. This vector is used in classification of face images using mahalanobis distance classifier. The method has been experimented on two standard databases: YALE consisting of RGB images and ORL consisting of grayscale images. The method found to be working very satisfactorily for RGB images over grayscale images. The maximum accuracy obtained through hysteresis class test is 81.43%.