Blind source extraction of long-term physiological signals from facial thermal images

Abstract

The amount of mental work using information equipment has been increasing because of a rapid growth of an information society. Accumulation of short-term mental work can cause various stresses and a disturbance of circadian rhythm and lead to fatigue, anxiety, and distraction. Estimation and understanding the physiopsychological states is desired for decreasing or controlling stresses to maintain health. There have been several investigations on the assessment of short-term physiopsychological states using infrared thermography. However, the method has been used rarely for assessing long-term physiopsychological states. In the present study, extraction of independent components related to long-term physiological signals is attempted by applying independent component analysis to facial thermal images obtained over 6 months (July–December). Furthermore, multiple regression analysis is attempted to create psychological model by facial thermal images. As the result, extracted independent components are shown to represent the strong features in nasal, mouth, cheek, eyebrow, and forehead regions. Attempting multiple regression analysis, features in nasal and mouth regions contributed to depression or dejection mood, features in cheek, eyebrow and forehead regions contributed to fatigue and features in tip of nasal, eyebrow and mouth regions contributed to state anxiety.