Abstract: A novel content-based image retrieval scheme using image edge informa
tion is proposed. Since color vector angles are only sensitive to hue and saturation but insensitive to luminance, edge information in a color image can be extracted based on them. A color vector angle histogram is obtained for edge pixels while a color histogram calculated for smooth pixels. Both histograms are jointly used to construct a characteristic vector of the image, which makes full use of color, edge and shape information in the image. Experimental results show that the proposed method is robust against luminance variations and the retrieval performance is better than those only using color histograms.

Key words: color vector angle, histogram, content-based image retrieval (CBIR)