Evaluating the segmentation result of a gray-level image

We propose in this communication an unsupervised criterion that enables to quantify the quality of a segmentation result of a gray-level image. The originality of this method lies in the possibility to evaluate the segmentation results for all kinds of images including textured ones. This method is based upon a new criterion that takes into account the intra-region and inter-regions disparities by considering the type of region (textured or uniform). These two disparity measures are computed from adaptive attributes calculated from the segmentation result. Experimental results show the efficiency of this technique for synthetic and natural images with textured regions.