Segmentation d'images par maximisation de l'entropie à deux dimensions basée sur le recuit microcanonique

In this paper, a new image segmentation method based on two-dimensional histogram analysis through entropy maximization is presented. To compensate for the weakness of the classical methods, that may be trapped into the first entropy local maximum met, a robust metaheuristic based on microcanonical annealing (MA) is introduced. The optimal segmentation thresholds are searched by looking only around the best configurations at different energy stages. Therefore the convergence is improved and the reproducibility of the optimal solutions is better guaranteed. The performance of the proposed method is illustrated through the segmentation of four biomedical images, and compared to the results obtained through Canny method.