Semantic Labeling and Multi-Modal Video Search
Professor Shih-Fu Chang
Tuesday, Nov 7, 11am
SITE 5084, 800 King Edward Ave
Distributed System Research Group
With the prevalence of visual content, researchers are no longer short of large quantity of image or video data in experiments. An even more encouraging condition is the availability of annotations and benchmarking resources. As a case in point, NIST TRECVID has offered unique data sets for evaluating advances in video indexing and retrieval. In its 6th year since inception, TRECVID currently provides more than 200 hours of news videos and labels of 400+ semantic concepts (location, objects, people, events, etc.) over 62,000 shots. As a separate data source, Web image search engines and photo forums also provide freely available but less reliable annotations of images. In this talk, I will describe some recent research exploiting the statistical classification models and the large data set to develop automatic techniques for generating probabilistic semantic indexes. I will present on-going research in combining such semantic indexes with other modalities like text from automatic speech transcription to improve video search, video event detection, and news topic tracking. At the end, I will discuss observations that no single modality satisfies needs of diverse types of video queries, and thereby motivate research of query-class-dependent multi-modal fusion models for effective video retrieval.
of Shih-Fu Chang
Shih-Fu Chang is a professor of electrical engineering and currently directs Digital Video and Multimedia Lab at Columbia University. He has made significant contributions in digital image/video search, adaptive video transmission, digital media forensics, and development of MPEG-7 international standards. His research has been among the most frequently cited in content-based image retrieval, with several large demo systems deployed such as VisualSEEk and VideoQ. He has been recognized with many prestigious awards, including a Navy ONR Young Investigator Award, an IBM Faculty Development Award, an ACM Recognition of Service Award, and a CAREER Award from the NSF. His group received a Best Paper Award from IEEE Transactions on Circuits and Systems for Video Technology in 2000, as well as several best student paper awards from IEEE and ACM. He served as a general co-chair of ACM Multimedia Conference 2000 and IEEE Multimedia Conference (ICME) 2004, a consultant for several media technology companies, and an advisor for several research institutions. He is an IEEE Fellow and currently Editor-in-Chief for IEEE Signal Processing Magazine.
website is maintained by Yongyi Mao, and was last updated on Oct. 1st,