Category Specific Face Recognition Based on Gender

Abstract

The idea of category specific face recognition is first to categorize facial images into categories based on visual cues like gender and race and then to perform face recognition using features specific to a category. One main problem in this approach is to find category specific features. We addressed category specific face recognition based on gender and explored which feature descriptor is more suitable for male category and which is more appropriate for female category. We tested four feature extraction techniques: Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), Local Binary Pattern (LBP) and Weber's Law Descriptor (WLD). To reduce the dimension of the feature space in each case, we used two-stage feature subset selection method. For classification, we used nearest neighbor classifier (NN), and tested the effect of different metrics on classification; spearman distance emerged to be the winner. The results showed that there is a trend that WLD is better feature descriptor for male category and LBP represents the female category in a better way.