Remote sensing image segmentation combining hierarchical Gaussian mixture model with M-H algorithm

To accurately model the complicated statistical distribution of pixel intensities with asymmetry, heavy-tailed and multimodal characteristics in a homogeneous region of high resolution remote sensing image, a high resolution remote sensing image segmentation algorithm combining hierarchical Gaussian mixture model (HGMM) with Metropolis-Hastings (M-H) algorithm was proposed. The HGMM, which was defined by the weighted sum of several Gaussian sub-mixture models (GSMM), was used to build the complicated statistical distribution model for high resolution remote sensing images. Following Bayesian theory, posterior distribution, namely the segmentation model was built by combining HGMM with the prior distributions of parameters. M-H algorithm was designed to simulate the segmentation model to segment image and estimate parameters. To verify the feasibility and effectiveness of the proposed algorithm, segmentation experiments on synthetic and high resolution remote sensing images were carried out. The experimental results were analyzed quantitatively and qualitatively. The results show that the proposed algorithm can accurately model the complicated distribution, and obtain results of high precision. © 2019, Editorial Board of Journal of CUMT. All right reserved.