Because of the weather- and illumination-independent characteristics, Synthetic Aperture Radar (SAR) has been playing a more important role for target recognition. Local stable feature descriptors in SAR image matching have been a interesting field in recent years. A new local feature extraction method like Scale Invariant Feature Transformation (SIFT) is proposed in this presentation, in which Local Gradient Ratio Pattern Histogram (LGRPH) based on SAR image similarity are taken as local feature descriptor from the neighbourhood of key points. Firstly, we extract the keypoints in difference of guassian (DoG) scale pyramid like many modified SAR-SIFT methods. Secondly, in the neighbour of kepoints, the local gradient ratio pattern histogram (LGRPH) is computed individually. Finally, the similarity is obtained by utilizing K-L discrepancy to measure the distance of LGRPH. Experimental results based on synthetic and real SAR images demonstrate that the proposed approach is robust to the speckle noise and local gradient variation in SAR images.