Accelerating multimedia search by visual features

Abstract

Visual features used for search of visual material usually have computationally complex similarity functions. Therefore for large databases to get real time response for queries by examples is necessary avoiding their full search. In this paper we show efficiency of selected techniques for accelerating visual object retrieval. They belong to three independent groups: filtering, partial similarity computing, and tree based data structures. We show on description examples of motion trajectory, face recognition, and distributed color image temperature that different types of visual features require different accelerating techniques. © Springer-Verlag 2004.