An effective image retrieval using the fusion of global and local transforms based features

Retrieval efficiency and accuracy are the important issues in designing content based image retrieval system. Thus, in this paper, we propose an effective descriptor based on angular radial transform (ART) and polar Hough transform (PHT), which is capable of fulfilling the above requirements. ART is used as a region based shape descriptor, which represents the global aspects of an image. PHT is used as a local shape descriptor for detecting linear edges in an edge image. The detected linear edges represent the association among adjacent edge points. The perpendicular distance of each linear edge to the centroid of the edge image is computed to build histograms, which exhibit rotation, scale and translation invariant properties. The combination of both global (ART based) and local (PHT based) features yields improved retrieval accuracy, which is analyzed over various sorts of image databases. An extensive set of experiments witness the superiority of the proposed hybrid system over other major contour based, region based and hybrid approaches.

► An efficient solution to image retrieval is proposed by applying ART and PHT.
► ART and PHT based global and local features are invariant to geometric transforms.
► The results of large set of experiments witness the primacy of the proposed system.