Body of water is a kind of important content of synthetic aperture radar (SAR) image interpretation. In this paper, a registration method, which is based on the contours and aiming at the imaging features of SAR image including body of water, is proposed. At first, an adaptive weighting Markov random field (MRF) segmentation model which is integrating local statistical information of observed image is proposed to segment the target of water body of SAR image and accurately extract its contour. Then, a non-uniform Gaussian mixture model (GMM) of contour matching is proposed. The mixture model can integrate both the location information of point of contours and the gray scale similarity information of windows including the contour points as the centers. At last, the registration experiments of SAR image including body of water are conducted. Results show that the proposed MRF segmentation model can accurately locate the edge of object and reserve the details of image. The non-uniform GMM for contours matching is robust to noise, outliers and local deformation, which can achieve the registration of SAR image including body of water better.