A geometry-enhanced color histogram

The color histogram, which bears the composition information of colors, has been widely used in content-based image retrieval. By further incorporating the spatial information, the effectiveness of conventional color histograms can be improved. However, the approaches reported in the literature usually demand high computation and storage cost. Here, we propose a geometry-enhanced method that incorporates the average distance of color pairs into the color histogram. The similarity between two geometry-enhanced color histograms is evaluated by simultaneously taking the composition and the distribution of colors into account. Experimental results show that the proposed geometric enhancement significantly improves the retrieval discriminability with minimal added complexity.