“…They commonly search for video shots by computing lowlevel features on entire or partitioned image frames, which are compared to those in consecutive frames to detect strong transitions (Lee, Yoo, & Jang, 2006;Zhu & Liu, 2009). Although low-level features are particularly useful for still image retrieval (Conci & Castro, 2002;Smeulders et al, 2000;Yoo et al, 2005) and video retrieval in movies, broadcast news, or sports (Xiong et al, 2006), they exhibit practical drawbacks for video surveillance.…”