Blind Remote Sensing Image Deblurring Based on Local Maximum High-Frequency Coefficient Prior and Graph Regularization

In satellite remote sensing imaging, factors such as optical axis shift, image plane jitter, movement of the target object, and Earth's rotation can induce image blur. The unavailability of the image fuzzy kernel makes the necessity for blind remote sensing image deblurring clear. This study introduces a priori constraint based on the maximization of local high-frequency wavelet coefficients in clean remote sensing images, integrated with a graph-based blind deblurring model. This approach aims to produce a skeleton image that retains sharp edge details while eliminating harmful structures, thereby enabling accurate estimation of the fuzzy kernel. An alternating iteration method, combined with a straightforward thresholding approach, is employed to address our proposed nonconvex, nonlinear model. Comparative experiments demonstrate that, relative to several leading blind image deblurring algorithms, our approach demonstrates unparalleled efficacy in enhancing peak signal-to-noise ratio and structural similarity index measurements.