Sea ice segmentation using Markov random fields

Tools are required to assist the identification of pertinent classes in SAR sea ice imagery. Texture models offer a means of performing this task. The texture information in SAR sea ice imagery can be characterized by two Markov random field models: the Gauss model for conditional distribution of the observed intensity image and the discrete model for the underlying texture label image. The segmentation can be implemented as an optimization process of maximizing a posteriori distribution in a Bayesian framework