Abstract:

Traditionally most of the video retrieval methods based on key words, so it is difficult to obtain satisfactory recall and precision. Therefore, we propose a video retrieval technique based on ontology, in which the basic concepts of domain ontology are set as key words to obtain sample image groups online via the web image search engine. SIFT features are extracted to establish the image features dictionary, and image feature histogram is extracted and its similarity is calculated to achieve automatic video annotation and initialize the database of video retrieval. Meanwhile, with the help of domain ontology, semantic extension of the key words extracted from the users' input is accomplished, then the retrieval results via the extended set of concepts are returned to the users, and the video retrieval based on ontology is realized.

Key words: domain ontology;video retrieval;semantic extension;automatic video annotation