Unsupervised SAR image segmentation using high-order conditional random fields model based on product-of-experts

摘要

Conditional random fields (CRF) model is suitable for image segmentation because this model directly defines the posterior distribution as a Gibbs field and allows one to capture the dependencies of the observed data. However, this model has a limited ability to capture the high-order label dependencies because of only the pairwise potential being constructed. Moreover, for synthetic aperture radar (SAR) image segmentation, SAR scattering statistics that are essential to SAR image processing are not considered in CRF model. Then for unsupervised SAR image multiclass segmentation, we propose a high-order CRF model based on product-of-experts (POE) in this paper, named as HOCRF-POE model. HOCRF-POE model decomposes the high-order label dependencies into the low-order ones and constructs the nonparametric high-order potential based on POE, thus effectively capturing high-order label dependencies. In addition, to capture SAR data information in a more completed manner in the unsupervised SAR image segmentation, HOCRF-POE model integrates the textural features and SAR scattering statistics under unsupervised Bayesian framework. The effectiveness of HOCRF-POE model is demonstrated by the application to the unsupervised segmentation of the simulated images and the real SAR images. (C) 2016 Elsevier B.V. All rights reserved.