A variation approach for SAR image feature enhancement is proposed. By analyzing the diffusion coefficient of P-M diffusion, it was found that it is the non-CFAR character of Canny detector that blurs the edges in the filtered SAR image. Further more, the ROA detector can detect edges in the image, but it lacks the ability of detecting the strong scatters. Based on the Rayleigh distribution of speckle in the amplitude image, the variation model for feature enhancement was established by combining the ROA detected version and the amplitude of SAR image from the MAP estimator. Experiments on the real SAR image demonstrated that the proposed method can suppress speckle efficiently in the homogeneous areas and preserve or even enhance edges and strong targets in the image.