Abstract
The conventional global color histogram bears only the color composition information of an image. Further incorporating color spatial information helps identify a color image. However, the joint spatial-chromatic relationship of real-world images has not been rigorously explored in the literature. We present a statistical model for the distance distribution of dominant colors. From an empirical evaluation of large image databases, we discover that the two-parameter Nakagami-m distribution captures well the spatial characteristics of colors. Over 90% accuracy with respect to the more complicated color autocorrelogram is achieved under this model