Abstract
More and more researches on the video retrieval express attribute changes and movements of the objects appearing in a video document. However, these expressions are hard to describe a complex “story” of a video document such as a movie. In order to query a video document according to the events and stories happening in it, a semantic video retrieval system is proposed. In this system, we use a hierarchical knowledge model to explain the relationships between each thing in the real world. This model is flexible and can be defined for different applications. Then we provide a video query language having many semantic characteristics. Users can use the terms of Noun, Verb and Relation to make a query that represents rich and complex semantic meanings. By cooperating with the proposed knowledge model, the retrieval system is able to do some inferences on the terms appearing in a query, and determine whether a candidate in the video database semantically matches them. Finally, a semantic similarity measurement is proposed to achieve the fault-tolerant query.