Feasible Human Emotion Detection from Facial Thermal Images

Abstract

Advances in sensing technologies have resulted in a broadening of their application areas. One typical application is sensing human status, mental as well as physical aspects. Almost all systems proposed to date need sensors to be attached to the subject’s body, an invasive requirement that increases the user’s burden. Therefore, we need noncontact and noninvasive methods. This paper verifies, experimentally, the possibility of detecting human emotion from facial thermal images taken by an infrared camera. In the experiment, two types of emotion, a somewhat positive feeling and a negative one, are verified by observing subjects performing two different tasks. Based on a preliminary experiment, we focus on the change in temperature of the nose. Thirteen young healthy males/females participated in the experiment. The results showed that the change in temperature depended on the task performed. The positive task yielded an increase (average of all subjects) of +0.52 °C, while the negative task triggered a −0.08 °C decrease. These results imply the method might be feasible. However, subjects exhibiting the cool finger effect showed a very small temperature variance. For noncool finger subjects, the measured temperature differences were +0.95 °C for the positive task and −0.36 °C for the negative task, which show that the method is feasible. Future work includes verifying the method with many subjects and using multiple parts of the face to increase recognition performance.