Facial Image-based Gender and Age Estimation

The principal objective of this project is to develop methods for the estimation of the gender and the age of a person based on a facial image, using classification for the gender estimation and regression for the age. The extracted information can be useful in, for example, security or commercial applications. This is a difficult estimation problem, since the only information we have is the image, that is, the looks of the person. Using image features such as gradients, pixel differences, and Histograms of Oriented Gradients (HOGs), and high-level features like hair, moustache or beard, a classifier/regressor is trained. The training process needs to be optimized in terms of pre-processing, feature selection, choice of classifier/regressor, and classification/regression parameters. Our experiments show that HOG is the most useful feature in order to estimate both age and gender. The Support Vector Machine is the best classifier and Random Forest is the best regressor. For gender estimation the misclassification rate is about 2%, and the performance of the age estimation is close to what humans achieve.