Face Recognition by Multi-resolution Curvelet Transform on Bit Quantized Facial Images

Multiresolution ideas notably the wavelet transform have been profusely employed for addressing problems in face recognition and other problems in computer vision. However, theoretical studies indicate that wavelets are not the ideal image descriptors; they have very crude directional representations and are not anisotropic. A recent development called the Curvelet Transform tries to overcome these shortcomings of wavelets. In this paper, we propose a multiresolution Curvelet based method for face recognition. Recognition will be done in three steps. In the first step the original 8 bit image is quantized to 4 bit and 2 bit representations. In the second step, each of the 3 different resolved versions of the image are subjected to Curvelet transform at five different resolutions. The approximate Curvelet coefficients at each resolution represent a different feature set. In the last step these fifteen (three bit resolved version x five resolutions) sets of coefficients are then used to train separate Support Vector Machines. During testing, the results of the fifteen SVMs are fused to determine the final result.