Search

Objectives: It is difficult to solve the matching problem between heterogeneous remote sensing images caused by nonlinear radiometric distortions. Methods: This paper proposes a nonlinear scale-space enhanced automatic matching method for optical and synthetic aperture radar (SAR) images. First, by modifying the calculation of color pixel contrast, the contrast information of images is effectively enhanced. As a result, the repeatability of corresponding points between optical and SAR images can be improved. Second, a nonlinear diffusion equation is employed to describe the image diffusion characteristics, avoiding the issue of boundary blurring in the Gaussian scale-space. Third, the multi-scale ratio of exponentially weighted averages operator and the Sobel operator are utilized to compute the gradient information of SAR and optical images, respectively, followed by the stable extraction of Harris feature points. Finally, log-polar de‑ scriptor framework is employed to compute a high discriminate feature vector, and the outliers are eliminat‑ ed by Euclidean distance and fast sample consensus algorithm. Results: The experimental results demonstrate that the proposed method can get more matching points and achieve higher matching accuracy, compared Objectives: It is difficult to solve the matching problem between heterogeneous remote sensing images caused by nonlinear radiometric distortions. Methods: This paper proposes a nonlinear scale-space enhanced automatic matching method for optical and synthetic aperture radar (SAR) images. First, by modifying the calculation of color pixel contrast, the contrast information of images is effectively enhanced. As a result, the repeatability of corresponding points between optical and SAR images can be improved. Second, a nonlinear diffusion equation is employed to describe the image diffusion characteristics, avoiding the issue of boundary blurring in the Gaussian scale-space. Third, the multi-scale ratio of exponentially weighted averages operator and the Sobel operator are utilized to compute the gradient information of SAR and optical images, respectively, followed by the stable extraction of Harris feature points. Finally, log-polar de‑ scriptor framework is employed to compute a high discriminate feature vector, and the outliers are eliminat‑ ed by Euclidean distance and fast sample consensus algorithm. Results: The experimental results demonstrate that the proposed method can get more matching points and achieve higher matching accuracy, compared with other classic methods. Conclusions: The proposed method can realize automatic and robust matching for SAR and optical images. [ABSTRACT FROM AUTHOR]