Content-Based Image and Video Retrieval
January 30, 2006
Prof. Sanjeev Khudanpur
Sanjeev Khudanpur is an Assistant Professor in the Department of Electrical & Computer Engineering and a member of the Center for Language and Speech Processing at Johns Hopkins University. He obtained a B. Tech. from the Indian Institute of Technology, Bombay, in 1988, and a Ph. D. from the University of Maryland, College Park, in 1997, both in Electrical Engineering.
We will describe the use of hidden Markov models (HMMs) for content-based retrieval of images and video via text queries. In this model, objects or concepts present in an image or video clip constitute the state-space of a Markov chain, and the observed visual features, as well as any text or caption accompanying the image, are modeled as stochastic emissions from the (unobserved) states of this Markov chain. A small set of manually annotated images are used to estimate parameters of the HMM. This estimated HMM then permits the
efficient computation of the posterior probability that any particular object or concept is present in any image or video clip in a large test (search) collection. Empirical results will be presented for the so called high-level feature detection task of the TRECVID 2005 benchmark test.