Combining structure, color and texture for image retrieval: A performance evaluation

In this paper we combine structure, color and texture for efficient image retrieval. Structure is extracted by the application of perceptual grouping principles. Color analysis is performed by mapping all pixels in an image into a fixed color palette that uses linguistic tags to describe color content. Texture analysis is done using a bank of even-symmetric Gabor filters. A methodology for performance evaluation of these analyses is presented on a database of color images. The database has been partitioned into various classes and subclasses for quantifying the success of image query and classification. It is demonstrated that the synergy resulting from the combination of structure, color and texture is superior than using just color and texture.