Robust multi-view UAV SAR image registration based on selective correlation of log gradient descriptor

Unmanned aerial vehicle (UAV) synthetic aperture radar (SAR) image characteristics, such as large geometric distortions, striped inhomogeneous radiation, severe speckle noise and ultra-high resolution, pose significant challenges to multi-view UAV SAR image registration. To address these issues, we propose a robust multi-view UAV SAR image registration algorithm based on the selective correlation of the log gradient descriptor. First, slant range UAV SAR images are geocoded using a publicly available digital elevation model to attenuate the topographic distortions. Second, we follow a template matching framework to obtain gridded correspondences. A log gradient that effectively handles noise in dark areas while resisting speckle noise is developed. A histogram of oriented log gradient (HOLG) is designed based on a simplified histogram of oriented gradient structure to weaken the effects of striped radiation noise. Moreover, a selective correlation strategy is designed, and a similarity metric named HOLGSC is proposed based on the selective correlation of the HOLG descriptors to overcome the effects of the unstable geometric structures of shadows and trees. Finally, after removing outliers from the matching results, we introduce the as-projective-as possible transformation model to establish geometric correspondences. We obtain several UAV SAR images and compose nine multi-view UAV SAR image pairs for experiments. The experimental results confirm that the proposed algorithm is effective and outperforms the existing state-of-the-art algorithms. In addition, a preliminary stitching result of these UAV SAR images is also presented. •This work proposes a robust multi-view UAV SAR image registration algorithm.•The algorithm includes a multi-scale extended log-based gradient.•The algorithm includes a novel descriptor with a simplified HOG structure.•The algorithm employs a new similarity metric using selective correlation.