Towards Thermal Region of Interest for Human Emotion Estimation

Thu Nguyen, Khang Tran, Hung Nguyen

Abstract: In the recent years, researches in implementing visual-based Automatic Facial Expression Analysis (FEA) and human emotions estimation have been a great interest and had widespread applications. However, vision-based visual systems face challenges such as poor-quality images due to low light conditions, poker-faces, the contrast of expressions and emotions. Because of the advantages of thermal images - unsensitive to ambient light conditions, we focus on developing a human emotion estimating system using thermal-based cues. It overcomes the limits of visible images, however, once a change in emotion occurs, only some areas of the face is affected in terms of temperature. In addition, the loss of thermal information in the glasses areas is a matter of concern. This paper presents our proposed method, thermal region of interest, to fill the gap when using thermal images to estimate seven emotions.