Contourlet-Based Feature Extraction with LPP for Face Recognition

Locality preserving projection (LPP) is a successful method in face recognition for feature extraction. However, the recognition efficiency of LPP technique is often degraded by the very high dimensional nature of the image space. It is difficult to calculate the bases to represent the original facial images. So the algorithm describing image in vector form is often applied in data after dimension reduction by PCA which result in the algorithm sensitive to how to estimate the intrinsic dimensionality of the nonlinear face manifold in the PCA preprocessing step. A novel approach is presented in this paper to avoid the difficulty. We introduce the application of contourlet transform in conjunction with LPP to overcome these limitations. Experimental results on the ORL, Yale, YaleB, CMU PIE face database show the effectiveness of the contourlet-based locality preserving projection (CLPP) method.