ELVIS: Entertainment-led video summaries

Abstract

Video summaries present the user with a condensed and succinct representation of the content of a video stream. Usually this is achieved by attaching degrees of importance to low-level image, audio and text features. However, video content elicits strong and measurable physiological responses in the user, which are potentially rich indicators of what video content is memorable to or emotionally engaging for an individual user. This article proposes a technique that exploits such physiological responses to a given video stream by a given user to produce Entertainment-Led VIdeo Summaries (ELVIS). ELVIS is made up of five analysis phases which correspond to the analyses of five physiological response measures: electro-dermal response (EDR), heart rate (HR), blood volume pulse (BVP), respiration rate (RR), and respiration amplitude (RA). Through these analyses, the temporal locations of the most entertaining video subsegments, as they occur within the video stream as a whole, are automatically identified. The effectiveness of the ELVIS technique is verified through a statistical analysis of data collected during a set of user trials. Our results show that ELVIS is more consistent than RANDOM, EDR, HR, BVP, RR and RA selections in identifying the most entertaining video subsegments for content in the comedy, horror/comedy, and horror genres. Subjective user reports also reveal that ELVIS video summaries are comparatively easy to understand, enjoyable, and informative.