An improved feature extraction techniques for gender classification and age estimation

Abstract

The face is a very significant biometric feature of humans. Gender classification and age estimation play a predominant role in the current era. Because of its growing real-world applications, it has obtained a lot of research and academic attention in recent decades. Currently, gender classification and age estimation have significant applications in various sectors such as Human-Computer Interaction (HCI), security control, surveillance monitoring, commercial development, content-based indexing and searching, demographic system, targeted advertising, biometric system, and forensic art. The necessity to improve existing methodologies leads the way for research in gender classification and age estimation research. Gender classification and age estimation are simple tasks for humans under any constraint. It is a challenging task for the machine to accurately identify gender and age due to pose variation, occlusion, illumination effect, facial expression, plastic surgery, and makeup. This thesis intends to present an effective feature extraction method to accurately identify gender and age under constraints like varying poses and facial makeup. Texture and shape features are considered for classifying gender under varying poses. The texture features are extracted using Dominant Rotated Local Binary Pattern (DRLBP) and Local Gabor Binary Pattern Histogram Sequence (LGBPHS) descriptors. The shape feature is extracted using Pyramid Histogram of Oriented Gradient (PHOG) descriptor. The experiments are carried out using Support Vector Machine (SVM) with different kernels for Adience, FEI, and Label Faces in the Wild (LFW) datasets. newline