Color image segmentation using a model-based clustering and a MFA-EM algorithm

摘要: In this paper we present a statistical model-based approach to the color image segmentation. A novel deterministic annealing EM and mean field theory are used estimate posterior probability of each pixel parameters Gaussian mixture model which represents multi-colored objects statistically. Image segmentation is carried out by clustering into most probable component Gaussian. The experimental results show that provides global optimal solution for ML parameter estimation real images segmented efficiently using estimates computed maximum entropy principle men theory.