A Fuzzy Theoretic Approach for Semantic Characterization of Video Sequences

Most of the existing video processing systems consider the problem of classifying the video sequences on the basis of syntactic features alone. These similarity measures may be adequate only if the goal is to nd frames with similar distribution of color, segmented regions or other low-level characteristics. We have worked on classi cation as a mechanism to support for indexing video sequences based upon intrinsic features of video data and/or their semantic content. First we segment video sequences into shots using a fuzzy framework. An opportunistic scheme is then used for processing the video sequences. Accordingly either mosaic-based or frame-based representation is used. We also compute various syntactic features like nature of cuts, camera motion, object motion and dynamics of scene. These features are evaluated for well de ned domains using fuzzy classi er. Domain-based semantic modeling and heuristic based object recognition are used to semantically de ne and recognize video sequences. Experimental analysis illustrates the e ectiveness of our system in o ering a novel approach for video classi cation.