Stochastic relaxation on partitions with connected components and its application to image segmentation

We present a new method of segmentation in which images are segmented in partitions with connected components. We give computationally inexpensive algorithms for probability simulation and simulated annealing on the space of partitions with connected components of a general graph. In particular, Hastings algorithms (1970) and generalized Metropolis algorithms are defined to avoid heavy computation. To achieve segmentation, we propose a hierarchical approach which at each step minimizes a cost function on the space of partitions with connected components of a graph. The algorithm is applied to segment gray-level, color, and textured images.