Robust SAR Image Registration Using Rank-Based Ratio Self-similarity
Xin Xiong, Guowang Jin, Qing Xu, Hongmin Zhang
Abstract
Synthetic aperture radar (SAR) images in different polarizations or from different sensors are becoming easily available, but registering these images is challenging because of the presence of significant speckles in SAR images and the existence of radiometric differences between images. To address the problems, we propose a novel feature descriptor named rank-based ratio self-similarity (RRSS) for robust SAR image registration. The descriptor first calculates a ratio surface by replacing the distance surface, as the use of the ratio is more robust to multiplicative noise, and then sorts the ratio values to construct the rank surface. Subsequently, the rank surface is partitioned into an index map, and the index map is then transformed into the descriptor vector based on a restricted adaptive binning grid to discriminatively describe features. Furthermore, a rotation invariance enhancement method is designed for the RRSS descriptor to efficiently calculate descriptor vectors in multiple orientations. We conduct experiments with six SAR image pairs of various bands, polarizations, and resolutions from different sensors, including ALOS-PALSAR, Gaofen-3, Sentinel-1, and TerraSAR-X. The results demonstrate that the proposed descriptor is superior to state-of-the-art descriptors and robust for SAR image registration.