Spatial segmentation of color images according to the MDL formalism

We present a new segmentation technique for multicomponent images. This technique is based on the minimum description length (MDL) formalism and aims at finding the best representation model of an image. The resulting segmentation algorithm permits to obtain good segmentations without the need of tuning parameters. Furthermore, the segmentation algorithm is invariant with respect to the linear space changes, which makes this algorithm well suited for color images where the choice of the color space is generally of great importance.