A Spontaneous Visible and Thermal Facial Expression of Human Emotion Database

In the last twenty years, analyzing facial expression and human emotion has received significant efforts among computer vision researches. Most of the works typically focus on categorizing emotions, but only few works address the interests in non-basic human emotion or dynamic facial expressions. Besides, existing databases are mostly based on visible images, which requires to be obtained under stable illumination condition. Therefore, in this study, a spontaneous human emotion database with various intensity in both visible and thermal spectrum is established. The database contains visible and thermal videos of 10 subjects generating 30 videos of facial behavior affected naturally to stimulus video clips with different intensities. A proposed analysis framework, then, is applied to assess the usability of our spontaneous database for nonbasic human emotion estimation. Firstly, a feature descriptor that represents well the relation between a single frame in thermal space and the dataset without losing discriminating abilities for subtle emotion variations is applied. Secondly, a ranking method that pays attention to the relative order information and takes a single thermal frame as input is employed in our assessment tool. The proposed database is available for research in computer vision, psychophysiology science and related fields