Remote Sensing Image Segmentation With Unknown Number Of Classes Based On RJMCMC Algorithm

In segmentation of remote sensing image, automatically determining the number of classes is a focus and difficult problem. To this end, an approach with unknown number of classes based on pixel and region using Reversible Jump Markov Chain Monte Carlo (RJMCMC) algorithm is presented. First of all, pixel-based and region-based image segmentation models are built using Bayesian paradigm. In a pixel-based image segmentation model, to model the correlation of pixels in terms of their labels, a Markov Random field (MRF) model is used. Assume that pixels in each homogeneous region satisfy an identical and independent statistical distribution. Then the Bayesian paradigm is followed to obtain the posterior distribution. To build a region-based segmentation model, an image domain is partitioned into a set of sub-blocks by regular tessellations. The label field is characterized by an improved Potts model. Assume that pixels in each block satisfy an identical and independent statistical distribution. The Bayesian paradigm is followed to build the image segmentation model based on the partitioned regions. In the RJMCMC algorithm, move types are designed, including splitting or merging real classes, updating parameter vector, updating label field, birth or death empty class and splitting or merging a block. To test the proposal approach, the segmentation is carried out on simulated images, Synthetic Aperture Radar (SAR) images and color remote sensing images. The testing results show that the proposed approach can not only determine the number of classes, but also segment homogeneous regions. Furthermore, the results of quantitative evaluation and qualitative evaluation show that the proposed approach works well and is very promising.