Palmprint Nonlinear Discriminant Feature Extraction and Recognition Based on Non-Subsampled Contourlet Transform

Palmprint recognition uses the principal lines of the palm skins to recognition applications. Compared with the face images, palm images have a high reliability and stability. Furthermore, the palm images are robust to the variation of the facial expression, post and illumination. The Non Subsampled Contourlet Transform has characteristics of good multiresolution, shift-invariance and high directionality. It can give an asymptotic optimal representation of edges and contours in image. Kernel discriminant analysis is an effective image feature extraction technique in the nonlinear algorithm. Based on these, we propose a kernel discriminant analysis based on non-subsampled Contourlet transform (NSCT) and palm recognition method. First, we apply the NSCT transformation to palm images and obtain new palm images with multiresolution and multidirectional. Then mapping the palm images to the kernel space, according to the kernel discriminant capacity to choose the new palm images with high discriminant capacities and use them to extract the discriminant features. Experimental results on Palm database show the effectiveness of the proposed method which has higher recognition performance compared with that based on the DCT, Gabor transformations.