Automated Estimation of Human Age, Gender and Expression

Recognition of facial variations has been a hot research area for the past decade. Age, gender and expression are three important facial variations attaining increasing attentions. This report presents a software system for automatic estimation of age, gender and expression. To make the system robust to any input image containing a face, a pre-processing stage is implemented to calibrate the face. Pre-processing stage also involves detection of eyes and nose. The age estimation subsystem is composed of aging feature extraction and feature classification. We use local binary patterns (LBP) and Gabor filter to describe the aging features. Feature classification is implemented based on histograms from LBP analysis. A minimum distance classifier is built upon training using images from FG-NET aging database. Furthermore, linear discriminant analysis (LDA) with fisher faces is used to identify gender and expression information. Experimental results based on training images demonstrate accuracies of 39.8%, 95.1% and 54.2% for age, gender and expression, respectively. Real-world testing also shows the estimation capability of this system. Keywords—Facial variations, age, gender, expression, automatic estimation, local binary pattern, linear discriminant analysis