Facial Expression Intensity Estimation Based on CNN Features and RankBoost

Abstract

Facial expressions provide a wealth of information that can help us understand a person's emotions and attitudes better. And the intensity of facial expression is very important for detecting and tracking the change of expression. In this paper, we present a frame work based on CNN features and Rank Boost algorithm to estimate the intensity of facial expression. In daily life, the change of facial expression is a process of dynamic changes over time. So the problem of estimating the intensity of facial expression can be converted into the sequencing problem of expression. The depth features based on deep learning have strong generalization ability. This paper utilizes the features obtained from CNN as input rather than the features from traditional machine learning. Further it enhances the ranking function of the weak hypothesis in the Rank Boost algorithm and adds more prior information into the loss function. Indeed, a large number of experiments on CohnKanade+ database show that the algorithm presented in this paper has better performance than the previous ones.