Nonsubsampled Contourlet Transform Based Descriptors for Gender Recognition

Gender recognition using facial images plays an important role in biometric technology. A key component of a gender recognition system is feature extraction. Motivated by the success of multiresolution techniques in various applications, we investigated four different feature extraction techniques based on Nonsubsampled Contourlet Transform (NSCT) to identify the best performing technique. We present a gender recognition system that uses SVM, two-stage feature selection and different feature descriptors based on NSCT. Among different NSCT based feature descriptors, the one based on NSCT and Weber Law Descriptor (WLD) gives the best accuracy (99.5±1.05) and it outperforms the state-of-the-art gender recognition systems on FERET database. This research reveals the best feature description technique using NSCT for gender recognition problem.