Reliable Age Estimation Based on Apt Gabor Features Selection and SVM

Abstract

Automatic and reliable facial image-based age estimation is becoming an intriguing research in computer vision and other related applications. The degrees of apparent visibility of the ageing process on a person’s face significantly differ from person to person, and thereby making facial image analysis based age estimation, a great challenge. In this paper, we propose a robust method for age estimation based on detailed feature analysis of the frontal facial image of a person. We employ the Gabor transformation based filters for feature extraction and implement exhaustive search to find the most appropriate orientations, scales and sizes of Gabor filters, for accurate age estimation using multi-class SVM classifiers. The deployment of conventional Adaboost classifier is also discussed to show the significance of our approach that involves exhaustive search. The proposed method is evaluated and compared on a standard FG-NET database according to the Leave-One-Person-Out (LOPO) protocol, and Mean Absolute Error (MAE), along with the Cumulative Score are adopted as quantification performance measures. The experimental results demonstrates the superior efficacy of the proposed method over many of the state-of-the-art methods under test.