Skip to Main Content
This paper presents an automatic segmentation approach for natural images based on adaptive mean shift and a statistical model-based approach. We employ the adaptive mean shift to determine the number of modes in a mixture model and to detect their components each mixture mode. We consider the use of the EM algorithm, combined with mean field annealing theory, for parameter estimation by Markov random field models from unlabelled data in the Gaussian mixture model (GMM). Lastly, the color images are segmented by using posterior probability of each pixel computed from the GMM. The experiment shows that the method can effectively and automatically segment natural images without specifying the number of initial components in GMM.