Highlight Detection in Movie Scenes Through Inter-users, Physiological Linkage

Abstract

Automatic summarization techniques facilitate multimedia indexing and access by reducing the content of a given item to its essential parts. However, novel approaches for summarization should be developed since existing methods cannot offer a general and unobtrusive solution. Considering that the consumption of multimedia data is more and more social, we propose to use a physiological index of social interaction, namely, physiological linkage, to determine general highlights of videos. The proposed method detects highlights which are relevant to the majority of viewers without requiring them any conscious effort. Experimental testing has demonstrated the validity of the proposed system which obtained a classification accuracy of up to 78.2%.