Robust t-mixture modelling with SMEM algorithm

Multivariate t-mixture modelling is more robust than Gaussian mixture modelling to a set of data containing a group or groups of observations with longer than Gaussian tails or a typical observations. To alleviate the problem of local convergence of the traditional EM algorithm, a split-and-merge operation is introduced into the EM algorithm for multivariate t-mixtures. The split-and-merge equations are first presented theoretically and then a new merge method is acquired. Accordingly, a modified EM algorithm is constructed. Experiments of data clustering and unsupervised color image segmentation are given.