Gender identification from frontal facial images using multiresolution statistical descriptors

Abstract

Gender identification is a significant task which is very useful in many computer applications like human-computer interaction, surveillance, demographic studies, and forensic studies. Being one of the most popular soft biometrics, gender information plays a vital role in improvement of the accuracy of biometric systems. In this paper, we have presented an approach based on multiresolution statistical descriptors derived from histogram of Discrete Wavelet Transform. First, the input facial imagewas enhanced by applying contrast limited adaptive histogram equalization. During feature extraction, multiresolution statistical descriptors were computed and fed into the NearestNeighbor, SupportVector Machine, and Linear Discriminant Analysis classifiers respectively.We have achieved encouraging accuracy for gender identification on complex dataset of frontal facial images.