Next:Face Recognition Up:Content Based Image Retrieval Previous:An Improved Appearance-based Approach

Hidden Markov Models for Scene Analysis

Users of image databases are often interested in finding a set of images that belong to a single identifiable linguistic class. An example of this is a search for so-called ``scapes'', such as landscapes, urban scenes, beach scenes, mountain scenes, seascapes, etc. We are employing Hidden Markov Models(HMM), which have been used very successfully in speech recognition but scarcely in computer vision, to analyze and interpret an image. An existing linguistic classification will be used, and learning methods will be employed to learn the probabilities associated with all nodes and transitions in the HMM.

D. Hernández, M.D. Levine

Annual Report

Mon Jun 26 21:22:20 GMT 2000