Multi-label annotation study in video semantic content analysis

• Published in: 2009 International Conference on Machine Learning and Cybernetics Vol. 4; pp. 2453 - 2457
• Main Authors: Wei Wei, Wen-Qing Liu, Li-Li Li
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2009
• Subjects: Computer science; Cybernetics; Information analysis; Information technology; Layout; Machine learning; Machine learning algorithms; Multi-label annotation; Semantic concept; Semantic scene annotation; Testing; Video content analysis; Video sharing; Videoconference
• DOI: 10.1109/ICMLC.2009.5212195
• ISBN: 9781424437023; 1424437024
• ISSN: 2160-133X

Annotation is an important step in video content analysis. In this paper, one inter-concepts strong association and dependency multi-label annotation method for video semantic concept is presented. In video content analysis process, concepts are often correlated. One concept in one shot are usually dependent on others concepts in the same shot. Co-occurrence of several semantic concepts could imply the presence of other concepts. Unlike previous approaches only to take into count the pair concepts correlations, the proposed methods exploits label correlations between concepts including more than three. For generation the inter-concepts association and dependency rules, join and prune techniques are employed to get potential semantic concept associations in one shot. Compound labels are considered as a single label in annotation step. Experiment results on real-world multi-label media data show that the performance of proposed method is relative satisfied.