Non-local spatially varying finite mixture models for image segmentation

In this work, we propose a new Bayesian model for unsupervised image segmentation based on combination of the spatially varying finite mixture models (SVFMMs) and non-local means (NLM) framework. The probabilistic NLM weighting function is successfully integrated into Gauss–Markov random field, yielding prior density that adaptively imposes local regularization to simultaneously preserve edges enforce smooth constraints in homogeneous regions image. Two versions our are proposed: pixel-based patch-based model, depending design function. Contrary previous methods proposed literature, approximation does not introduce parameters be estimated because completely known once neighborhood pixel fixed. can closed-form solution via maximum posteriori (MAP) estimation an expectation–maximization scheme. We have compared with previously SVFMMs using two public datasets: Berkeley Segmentation dataset BRATS 2013 dataset. performs favorably approaches achieving better results terms Rand Index Dice metrics experiments.