Segmentation Evaluation Using a Support Vector Machine

Segmentation evaluation is a very difficult task even for an expert. We propose in this article a new unsupervised evaluation criterion of an image segmentation result. The quality of a segmentation result is derived without any a priori knowledge by taking into account different evaluation criteria from the literature. We first compare 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. We propose in this paper to fuse the best ones by a support vector machine. We illustrate the efficiency of the proposed approach through some experimental results.