In this paper, a novel automatic change detection approach in multi temporal SAR(synthetic aperture radar) images is proposed. This method mainly consists of the following two models. In model 1, a difference image is obtained by log ratio algorithm, and in model 2, a difference image is divided into three parts: changed, unchanged and uncertain. Markov random field which is used in multi temporal SAR images change detection is feasible. This approach, unlike classical ones, spatial contextual information in the neighborhood is contained and change detection results are much more accurate compared to classical ones. Experimental results from multi temporal SAR images of some region of Dongtinghu Lake, Hunan Province, confirm the validity of the proposed approach.