Automatic Video Intro and Outro Detection on Internet Television

Abstract

Content Delivery Networks aim to deliver multimedia content to end-users with high reliability and speed. However, the transmission costs are very high due to large volume of video data. To cost-effectively deliver bandwidth-intensive video data, content providers have become interested in detection of redundant content that most probably are not of user's interest and then providing options for stopping their delivery. In this work, we target intro and outro (IO) segments of a video which are traditionally duplicated in all episodes of a TV show and most viewers fast-forward to skip them and only watch the main story. Using computationally-efficient features such as silence gaps, blank screen transitions and histogram of shot boundaries, we develop a framework that identifies intro and outro parts of a show. We test the proposed intro/outro detection methods on a large number of videos. Performance analysis shows that our algorithm successfully delineates intro and outro transitions, respectively, by a detection rate of 82% and 76% and an average error of less than 2.06 seconds.