Video clip representation and recognition using composite shot models

This paper proposes using composite shot models to represent and recognize short video clips, such as advertisement clips. The temporal constraints of the shots constituting the clip are represented by a directed graph, which takes the shots as the nodes. Because each shot is represented by a hidden Markov model, the whole clip representation becomes a composite HMM. The test sequence is recognized by computing its likelihood belonging to the composite HMM. The experiment results show that composite HMM is not only robust to feature distortion induced by transcoding, but also demonstrates quite good performance on recognizing shorter versions. This representation scheme is also a general and flexible framework, which allows the choice of shot representation, and different shot representations should have different matching methods.