Extracting the oil spills in SAR images has been challenged by the spatial heterogeneity caused by the speckles. In this paper, we present a novel variational formulation based on the local characteristics of SAR images for CV mode. The proposed method is aimed to solve the segmentation problem caused by SAR image speckles. Compared with the traditional CV model, the performance of the improved CV model is verified by plenty of real airborne SAR images and the experimental results on the real data show its efficiency and accuracy. By presenting various results in the classical threshold segmentation method and Markov random field method respectively, we conclude that the improved CV model is effective and accurate.