Discrimination of Stress Coping Style Based on Facial Thermal Images Applying t-SNE and Sparse Modeling

When exposed to external stimuli, people show two types of responses. These are called stress coping responses or active and passive coping responses. These stress coping responses are discriminated by differences in the fluctuations of hemodynamic parameters such as cardiac output (hereinafter referred to as CO), total peripheral resistance (hereinafter referred to as TPR), mean blood pressure (hereinafter referred to as MBP), and others. However, the existing measurement method for hemodynamic parameters is contact measurement, and a remote measurement method is required for application. Therefore, we focused on Facial thermal image, one of the remotely measurable indicators of the cardiovascular system. We have attempted to construct a model to estimate stress coping responses from spatial characteristics of Facial thermal images using sparse coding. However, those studies had problems in terms of low accuracy and individual differences. Therefore, in this study, we attempted to construct a model that takes into account the distribution of facial thermal images during stress coping responses. In the experiment, to elicit each stress coping response, a cold pressure task and a game task were performed. Facial thermal images and hemodynamic parameters were measured during the experiments. The measured hemodynamic parameters confirmed the elicitation of a stress coping response. Also, the measured facial thermal images were mapped into a two-dimensional space by using a dimensionality reduction technique. As a result, the distribution of facial thermal images in each stress state was confirmed. In addition, based on the distribution, facial thermal images were selected for each stress state, and a stress coping style discrimination model was reconstructed. As a result, the accuracy was improved.