Gender Recognition for Juvenile Unconstrained Faces Using Gabor-MeanPool-DCT Feature Model and SVM-Kernel Optimization

Abstract

This paper presents novelty for facial gender analysis in unconstrained pose for children face where feature of gender categorization are immature. Adult face has matured feature for gender specification while face of juvenile age of group 2 to 14 year is very hard to recognize by system through digital face only. Face image having variation in illumination, pose (position, orientation, scale and expression) and obstructions as cuts, moles, eye color, birth marks, cap, scarves, bandage, spectacles etc. makes more hard to classify juvenile face in attribute of male or female. The paper contribute in feature engineering for illumination invariant, compact feature extraction for juvenile faces in unconstrained environment. Juvenile face has uniqueness feature in edges and texture pattern for different gender category. Gabor filter is illumination invariant and generate edge, textural features but redundant feature coefficients and huge dimensions. In the proposed method, the problem is solved using MeanDCT to generating compact and unique features for efficiency and accuracy respectively.