| Summary: | 博士 === 國立清華大學 === 資訊工程學系 === 93 === In this dissertation, a video data model is proposed to represent the content of video data. The perceptual and semantic properties of the video objects appearing in the video scene are recorded to represent the content of the video scene. For different applications, a predefined hierarchical knowledge model is used to express the semantic meanings contained in the videos such as the event happening in the video or the relationship between two objects in the video. In the proposed model, the trajectory and other properties of objects are recorded. From the trajectory, the motion events such as “high speed” of an object and “increasing distance” between objects can be automatically derived. Moreover, the semantic properties such as the actions performed by the object and the relationship between video objects are also recorded in the proposed model. A query language named V-SQL based on the video data model is also proposed for the users to describe the content of the desired video clips. The objects, the actions performed by the objects and the relations among objects are used to specify rich and complex semantic meanings in a query.
The concept hierarchy is used for user to specify concept query. The index structure, including the buckets for the attribute, suffix tree for motion sequence of video objects and the metadata table of the semantic properties of the video scene, is constructed to retrieve the matched video object based on the selected element in the concept hierarchy. The user specified concept query as well as the motion query will be processed through the index structure to find the corresponding video objects. Based on the proposed video model, the approximation query is also considered to find the similar retrieval result. Based on the proposed knowledge model, the retrieval system is able to make inferences on the predicates in a query, and determine whether a video semantically matches the query conditions. The semantic similarity measurement is also proposed for processing approximate queries.
Video data can be represented as a multiple-attribute string of feature values corresponding to multiple features of the data. Therefore, the retrieval problem can be transformed into the q-attribute string matching problem if q features are considered in a query. Traditional string matching algorithms cannot be efficiently applied to the q-attribute string matching problem. In this dissertation, two index structures and the corresponding matching algorithms, which can be applied on different values of q, are proposed for exact and approximate q-attribute string matching problem. The performance analysis and the experiment results are presented to show the efficiency of the proposed algorithm.
|
|---|
