Gender classification across different age groups: An extensive benchmark evaluation

Abstract

Gender is a non-intrusive, soft biometric trait which is gaining popularity in the field of biometric authentication systems for making them more robust. Gender classification across different age groups becomes challenging when it is carried out using facial characteristics, which varies due to photometric reflectance properties of human skin. In this paper, we carried out extensive study using the visible facial images, across 11 different state-of the-art feature extraction methods employed in gender classification followed by linear Support Vector Machine (SVM) classifier across different age groups. Experiments are performed independently on visible face database comprising of 313 subjects of which 183 are male and 130 are female from both Adults and Child age groups. The empirical performance of study obtains higher average classification accuracy of 94.843.35 percent for age groups greater than or equal to 20 years i.e Adults using Log Gabor (LG) feature extraction method. While the average classification accuracy for age group less than or equal to 15 years i.e Child was 89.744.57 percent using Binarized Statistical Image Features (BSIF), which indicates the variations in the performance accuracy for two different age group data.