Two-Step Matching Approach to Obtain More Control Points for SIFT-like Very-High-Resolution SAR Image Registration

Abstract

Airborne VHR SAR image registration is a challenging task. The number of CPs is a key factor for complex CP-based image registration. This paper presents a two-step matching approach to obtain more CPs for VHR SAR image registration. In the past decade, SIFT and other modifications have been widely used for remote sensing image registration. By incorporating feature point location affine transformation, a two-step matching scheme, which includes global and local matching, is proposed to allow for the determination of a much larger number of CPs. The proposed approach was validated by 0.5 m resolution C-band airborne SAR data acquired in Sichuan after the 2008 Wenchuan earthquake via a SAR system designed by the IECAS. With the proposed matching scheme, even the original SIFT, which is widely known to be unsuitable for SAR images, can achieve a much larger number of high-quality CPs than the one-step SIFT–OCT, which is tailored for SAR images. Compared with the classic one-step matching approach using both the SIFT and SITF–OCT algorithms, the proposed approach can obtain a larger number of CPs with improved precision.

GPU Accelerated Processing Method for Feature Point Extraction and Matching in Satellite SAR Images

“…Wang et al [19] introduced a SAR-SIFT algorithm based on the Range-Doppler (RD) model, incorporating the RD model into feature matching and geometric transformation estimation, and designing a novel local matching approach based on the RD model. Deng et al [20] presented a two-step matching method based on SIFT, utilizing global matching to establish coarse mapping relationships and employing a refinement matching strategy for precise matching of SAR images.…”