Searching segments of interest in single story web-videos

This paper presents a method for predicting the parts of a video that could be marked as “interesting” by an user. Our approach consists in considering the three competing crite-ria: salience, expressivity and significativity and to automatically combine the three corresponding function-of-interest. We evaluate this system on an user-annotated test set. Results demonstrate that, in spite of the intrinsic subjectivity of user choices, the system succeeds in finding about 51% of interesting segments.