Skip to Main Content
In this paper, we present a split-merge clustering segmentation algorithm based on Gaussian mixture models, which resolves the models by expectation-maximization (EM) algorithm and seeks model via Bayesian information criterion (BIC). It starts iteratively splitting from a single Gaussian model, then iteratively merging clusters. After convergence of the last stage, the clustering model is selected via a modified BIC and used to gain an initial segmentation, followed by a region merge step to achieve final segmentation. New algorithm was applied to 60 range images acquired by two kinds of range cameras, and got approving results with acceptable computation time.