Segmentation evaluation by fusion with a genetic algorithm

The goal of this work is to be able to quantify the quality of a segmentation result without any a priori knowledge. We propose in this article to fusion different unsupervised evaluation criteria. In order to identify the best ones to fusion, we compared six unsupervised evaluation criteria on a database composed of synthetic gray-level images. Vinet's measure is used as an objective function to compare the behavior of the different criteria. A new criterion is derived by linearly combining the best ones. The linear coefficients are determined by maximizing the correlation factor with the Vinet's measure by a genetic algorithm. We present in this article some experimental results of evaluation of natural gray-level images.