Angular-Radial Decomposition Algorithm for Sketch-Based Image Retrieval

This paper presents a novel algorithm for sketch-based image retrieval. The approach enables measuring the similarity between a full color multi-component model image and a black and white sketched query and needs no cost intensive image segmentation. Two different procedures, based on strong edges of the model image and thinned outline of the sketched image, are exploited to derive two abstract images. Angular-radial decomposition of pixels in the abstract images is used to extract new compact and effective features. A collection of paintings and sketches (ArT BANK) is used for testing the proposed method. Average Normalized Modified Retrieval Rank (ANMRR) is used for performance evaluation. The results are compared with three other well-known approaches. Experimental results show significant improvement in ANMRR factor using the proposed features.