A multiphase level set clustering approach using MRF-based Student's-t mixture model

Student's-t distribution has attracted widely attention on model-based clustering analysis. In this paper, we propose a new level set energy function framework where the Markov random field-based Student's-t mixture model is incorporated for clustering both static images and time-series data. This algorithm provides a general strategy by taking the best of Bayesian technique and level set formulation. A remarkable advantage of the proposed method is that it can overcome the weakness of the classical level set method by filtering out the outliers and stopping at the boundary points. It is mainly because the proposed technique models the probability density function of the data via Student's-t mixture model. Another attractive feature is that the local relationship among neighboring pixels is considered into mixture model so that the proposed framework is more robust against noise compared to the other level set based models. Expectation maximization algorithm is applied to obtain model parameters by maximizing the log-likelihood function. Additionally, the proposed model has simplified structure which sharply reduces the computational complexity. Finally, numerical experiments on various synthetic, real-world images, and time-series data are conducted. The performances are compared to other related approaches in terms of effectiveness and accuracy.