Minimization of color image entropy through colorimetric transformation

Image compression plays an important role in making digital images useful. Good compression systems must provide a trade-off between quality of representation for human beings and bit rate. We examine how entropy depends on choice of color space. We study entropy in both theoretical distributions of color image RGB values and distributions drawn from natural images. We determine which of several color spaces has minimal entropy, including several that purport to model human color processing. Our analytical results are supported by results found with a variety of images. Both analytical results and empirical results show that the UVW color space best minimizes entropy.