maybe it would be better to weight pixels by the energy density of the image region so that backgrounds (sky, walls) don't dominate the resultant colors over the (usually more significant) foreground/detailed objects