Age Estimation Using Bayesian Process

Abstract

Age problems have attracted many researchers’ attentions in recent years since they have many potential applications in human-computer interaction and other areas. Among all the age problems, automatic age estimation is one interesting problem and many methods have been proposed to solve this problem. In this paper, we use two Bayesian process regression algorithms, Gaussian process and t process, for age estimation. Different from previous regression methods on age estimation, which need to specify the form of regression functions or determine many parameters in regression functions in inefficient ways such as cross validation, in our methods, the form of regression function is implicitly defined by kernel function and almost all the parameters of our methods can be learnt from data automatically using efficient gradient methods. Moreover, our methods are very simple and easy to implement. Since Gaussian process is easy to be affected by outlier data points, t process can be viewed as a robust version of Gaussian process to solve this problem. Experiments on one public aging database FG-NET show our method is effective and comparable with the state-of-the-art methods on age estimation.