Distance measures in RGB and HSV color spaces

Abstract

This paper provides a comprehensive study of the different distance measures proposed in literature for RGB and HSV color spaces. This paper provides a reference for all those working on color similarity. Two key problems in color image analysis are (1) determination of a suitable color space and (2) finding the best distance measure. Color is perceptually divided into chromaticity and intensity components, where the former capture the perceptual notion of color and the latter, the harder to define notion of brightness or luminosity. A major problem of many existing distance functions is that identical weights are assigned to all the components in their respective coordinate systems. Thus, these do not match human perception. Two new distance measures are proposed, which discriminate two colors with greater emphasis on their chromaticity than luminosity, thus better matching human perception. Our weighted LI HSV and geodesic HSV distances are compared against the other distances on a test image set. We discuss the application of these distances in forensic document examination, where a significant amount of control and flexibility is necessary in analyzing colors. We apply them to content based image retrieval and initial results show that these distances improve performance compared to the popular Euclidean distance.