Photoplethysmography-Based Emotion Recognition in Response to High Dynamic Range Videos

Emotions are complex subjective states that exhibits a strong correlation with physiological or bodily, cognitive process and behavioral changes. Identifying and understanding human emotions holds significant importance as it incorporate wide range of application i.e. virtual reality (VR), Human computer interaction (HCI) and in developing healthcare system. Previously, affective computing have been carried out using traditional multimedia content i.e., standard dynamic range (SDR) videos, speech and images. High dynamic range (HDR) content producers aim at delivering more immersive and realistic visual experience and Quality of Experience (QoE) via wider range of luminosity, contrast and saturation level which meets the perception capability of human visual system (HVS) in real world. In this paper, PPG signals that were recorded from 27 participants in response to four HDR videos are analyzed. Each video was selected based on its mapping of target emotion in the arousal-valence space, aiming to serve as emotionally evocative stimuli. Time domain features are extracted and selected from the prepossessed signals to classify emotions in two arousal and valence classes using support vector machine (SVM) classifier. Classification accuracy of 78.7% and 63.8% has been achieved for arousal and valence, respectively. Our findings reveals that HDR content can powerfully stimulate the strong emotions.