Towards exploring RGB color channels of visible face images to band for gender classification

Abstract

Gender classification across different spectra has consistently been a challenging task due to the spectral gap between the training and testing samples. This work explores the Red (R), Green (G) and Blue (B) channel independently for visible to spectral band gender classification, leveraging photometric normalization methods and Probabilistic Collaborative Representation Classifier (ProCRC). We employ 22 photometric normalization methods to address the spectral gap between training and testing samples. We investigate the experimental evaluation result separately on R, G, and B channels based on 145 subjects which corresponds to visible and multi-spectral face database. We report the average classification accuracy by conducting the experiment over 10 different trials and the highest classification accuracy reported is 95.75 plus or minus 1.87 percent demonstrating the applicability of our approach for gender classification.