Using perceptually weighted histograms for colour-based image retrieval

In common colour-based image retrieval, colour histograms for images in the database and queries are calculated. The distance between a query and each of the database images is calculated as the sum of the absolute values of bin-to-bin differences between their histograms. The method ignores colour similarity between bins, leading to cases where perceptually similar images have very large histogram distances. In this paper, we propose to use perceptually weighted histograms (PWH) to overcome the problem. In PWH, a pixel contributes weights to a number of perceptually similar bins instead of a single bin. The contributing weights are inversely proportional to the distance between the pixel and bins.