Age classification with deep learning face representation

Automatic age estimation from facial images is challenging not only for computers, but also for humans in some cases. Therefore, coarse age groups such as children, teen age, adult and senior adult are considered in age classification, instead of evaluating specific age. In this paper, we propose an approach that provides a significant improvement in performance on benchmark databases and standard protocols for age classification. Our approach is based on deep learning techniques. We optimize the network architecture using the Deep IDentification-verification features, which are proved very efficient for face representation. After reducing the redundancy among the large number of output features, we apply different classifiers to classify the facial images to different age group with the final features. The experimental analysis shows that the proposed approach outperforms the reported state-of-the-arts on both constrained and unconstrained databases.