Segmentation Of Brain Mr Images Using J-Divergence Based Active Contour Models

Abstract

In this chapter we propose a novel variational formulation forbrain MRI segmentation. The originality of our approach is onthe use of J-divergence (symmetrized Kullback-Leiblerdivergence) to measure the dissimilarity between local andglobal regions. In addition, a three-phase model is proposed toperform the segmentation task. The voxel intensity value of allregions is assumed to follow Gaussian distribution. It isintroduced to ensure the robustness of the algorithm when animage is corrupted by noise. J-divergence is then used tomeasure the ``distance'' between the local and global regionprobability density functions. The proposed method yieldspromising results on synthetic and real brain MR images.