WEIGHTING SPATIAL ARRANGEMENT OF COLORS IN CONTENT BASED IMAGE RETRIEVAL

Content-based access to image databases relies on models which permit one to abstract images in some space of visual features. A large number of such models, addressing shape, texture, color and spatial arrangement features, have been proposed and experimented with. The fusion of multiple features within a single model is now being investigated as a way to increase the effectiveness of retrieval. In particular, the integration of color and spatial models is receiving growing interest. In this paper, we propose a representation which captures the spatial arrangement of the color sets of a picture, without talking into account their distribution in separate non-connected regions. A spatial modeling technique, which we call "weighted walkthroughs", is presented, and its suitability to support this representation is motivated and experimented on.