Analysing user physiological responses for affective video summarisation

Abstract

Video summarisation techniques aim to abstract the most significant content from a video stream. This is typically achieved by processing low-level image, audio and text features which are still quite disparate from the high-level semantics that end users identify with (the ‘semantic gap’). Physiological responses are potentially rich indicators of memorable or emotionally engaging video content for a given user. Consequently, we investigate whether they may serve as a suitable basis for a video summarisation technique by analysing a range of user physiological response measures, specifically electro-dermal response (EDR), respiration amplitude (RA), respiration rate (RR), blood volume pulse (BVP) and heart rate (HR), in response to a range of video content in a variety of genres including horror, comedy, drama, sci-fi and action. We present an analysis framework for processing the user responses to specific sub-segments within a video stream based on percent rank value normalisation. The application of the analysis framework reveals that users respond significantly to the most entertaining video sub-segments in a range of content domains. Specifically, horror content seems to elicit significant EDR, RA, RR and BVP responses, and comedy content elicits comparatively lower levels of EDR, but does seem to elicit significant RA, RR, BVP and HR responses. Drama content seems to elicit less significant physiological responses in general, and both sci-fi and action content seem to elicit significant EDR responses. We discuss the implications this may have for future affective video summarisation approaches.