A new local feature extraction in SAR image

Local feature extraction and matching is an essential tool for image registration and object recognition. In this presentation we present a novel local feature extraction method in SAR image. The extraction method includes three parts like SIFT: (i)keypoints detection and localization; (ii)keypoint description; (iii)keypoints matching. But our descriptor is based on local gradient ratio pattern histogram rather than gradient direction histogram in SIFT, which is more robust against speckle in SAR image. This novel method shows better performance than other algorithm in experiments using real SAR image.