Structural information, on the form of shot patterns and timing, plays an important role in scripted video, being consistently used by filmmakers as a way to convey meaning and significance into their work. For this reason, a video retrieval system can benefit from exploiting this kind of information. In this paper we propose a novel method for the retrieval of video scenes according to their structural similarity, based on graph-theoretical measures and vector quantization techniques. We show the results of our method in some scene retrieval experiments on a data set of 802 movie scenes, extracted from a set of 180 mainstream movies.
展开▼
