Change Detection in Color Images

One approach to change detection that is insensitive to illumination brightness and spectral changes uses properties of color imagery. This approach does not appear to have been previously investigated, and this paper asks: “What is the most appropriate metric for detecting changes in color imagery?”. We define and compare six image difference functions, mainly based on RGB and HSV image representations. When the global illumination does not change radically, the normal Euclidean distance in the RGB or a modified HSV space works about equally well, when there is significant changes in image spectrum, then HSV works a bit better and if there is a large local change in the illumination, none of the investigated methods work