Abstract
Face recognition performance can be improved when face images are first classified into categories and then analysed with category-specific descriptors. One such category is gender. The face image is a type of texture that can be represented using texture descriptors. We employ two state-of-the-art texture descriptors, the local binary pattern (LBP) and Weber's law descriptor (WLD), and investigate their spatially enhanced versions (SLBP and SWLD) for gender classification. A suitable choice of parameters used in these descriptors leads to significant improvement. The best combination of parameters is found through a large number of experiments performed on the FERET and Multi-PIE databases. Using these parameters, the SLBP and SWLD perform much better with less algorithmic complexity compared to commonly used gender recognition approaches.