Unsupervised evaluation of image segmentation application to multi-spectral images

We present in this article a study of some unsupervised evaluation criteria of an image segmentation result. The goal of this work is to be able to automatically choose the parameters of a segmentation method best fitted for an image or to fuse different segmentation results. We compared six unsupervised evaluation criteria on a database composed of 100 synthetic gray-level images segmented by four methods. Vinet's measure is used as an objective function to compare the behavior of the different criteria. We finally apply these criteria to evaluate segmentation results of multi-components images. We present in this article some experimental results of evaluation of gray-level and multi-components' natural images.