To address the problem caused by speckle noise in SAR image feature extraction, a method based on nonlinear scale space is proposed, to extract scale invariant feature points. First, instead of down sample in scale space construction, multilook process is utilized to build multisolution images, which improves radiometric resolution and reduces speckle noise. Then, a nonlinear scale space is constructed based on a novel enhanced Lee filter, which preserves feature details as well as reducing speckle noise further. The experiment shows that the proposed method is not only perform better in speckle noise suppression, but also increase the amount of feature points and amount of correctly matched points.