Combining adjacent self-similarity feature and enhanced matching method for sentinel image registration

The accurate registration of Sentinel-2 optical images with Sentinel-1 synthetic aperture radar (SAR) images plays a critical role in remote sensing applications such as disaster management, environmental monitoring and military target identification. The distinct imaging mechanisms of the two images result in non-negligible radiometric and geometric differences, which make traditional image registration techniques suffer from insufficient correct matching points or even registration failures. To solve this problem, an optimised image registration algorithm is proposed. Firstly, First, the structural features are obtained by computing the adjacent self-similarity (ASS) features, and effectively suppress the interference of radiometric differences and speckle noise on the registration process through constructing index map and minimum self-similarity map (SSM). Subsequently, a new key-point detector is designed to restrain the local non-maximum of the minimum SSM, and a descriptor for the distribution histogram is built by log-polar grids. Finally, we employ a cascaded sample consensus (CSC) matching method that refines the search region for correct matches, leading to high accurate correspondences. In summary, our algorithm offers a comprehensive solution for image registration that improving both the number of correctly matched key-points and the overall registration accuracy, and is applicable to any complex scenario.