Strategies of Shape and Color Fusions for Content Based Image Retrieval

The aim of this paper is to discuss a fusion of the two most popular image features - color and shape - in the aspect of content-based image retrieval. It is clear that these representations have their own advantages and drawbacks. Our suggestion is to combine them to achieve better results in various areas, e.g. pattern recognition, object representation, image retrieval, by using optimal variants of particular descriptors (both, color and shape) and utilize them in the same time. To achieve such goal we propose two general strategies (sequential and parallel) for joining elementary queries. They are used to construct a system, where each image is being decomposed into regions, basing on shapes with some characteristic properties - color and its distribution. In the paper we provide an analysis of this proposition as well as the initial results of application in Content Based Image Retrieval problem. The original contribution of the presented work is related to the fusion of several shape and color descriptors and joining them into parallel or sequential structures giving considerable improvements in content-based image retrieval. The novelty is based on the fact that many existing methods (even complex ones) work in the same domain (shape or color), while the proposed approach joins features from different areas.