Robust Gender Classification Using Multi-Spectral Imaging

Abstract

Multi-Spectral imaging is gaining importance in recent times due to it's ability to capture spatio-spectral data across the electromagnetic spectrum. In this paper, we present a robust gender classification approach by exploring the inherent properties of multi-spectral imaging sensor. We propose a framework that processes the spectral data independently using Spectral Angle Mapper (SAM) and Discrete Wavelet Transform (DCT), which are further combined to learn in a linear Support Vector Machine (SVM) classifier, the gender prediction. We present an extensive set of experimental results in the form of average classification accuracy using multi-spectral face database of 78300 samples images corresponding to 145 subjects in six different illumination conditions. The highest average classification accuracy of 96.80±1.60% is obtained using proposed approach signifying the potential of multi-spectral imaging for robust gender classification.