Smooth Image Segmentation by Nonparametric Bayesian Inference

Peter Orbanz, Joachim M. Buhmann

Abstract: A nonparametric Bayesian model for histogram clustering is proposed to automatically determine the number of segments when Markov Random Field constraints enforce smooth class assignments. The nonparametric nature of this model is implemented by a Dirichlet process prior to control the number of clusters. The resulting posterior can be sampled by a modification of a conjugate-case sampling algorithm for Dirichlet process mixture models. This sampling procedure estimates segmentations as efficiently as clustering procedures in the strictly conjugate case. The sampling algorithm can process both single-channel and multi-channel image data. Experimental results are presented for real-world synthetic aperture radar and magnetic resonance imaging data.