Video Semantic Models : Survey and Evaluation*

Summary With the development of video technology and appearance of new video-related applications, the amount of video data has increased dramatically which demands support in semantic models to facilitate information representation and query. The video semantic models surveyed in this paper are classified into two categories: annotation-based models and rich semantic models. However, currently there are no criteria for a good semantic model so people lack the rules for evaluating an existing model and the guidelines for designing a new model when necessary. To address this issue, this paper proposes twenty one properties as the criteria for video semantic models, and evaluates eleven existing rich semantic models according to these properties. The results show that these models mostly concentrate on aspects of basic expressive power and query ability. But for some advanced features such as user-defined constraints, assistance for acquisition of semantic information, query evolution etc., there are rooms for further enhancement. The paper concludes by analyzing the evaluation results and indicating research directions for future video semantic models.