Answers are grounded in this paper and cite their sources.
Multi-scale stripe noise and blur degradation in wide-field infrared remote sensing images are strongly coupled, posing significant challenges for image restoration. Most existing approaches rely on sequential processing frameworks, in which destriping and deblurring are performed independently. However, residual artifacts or over-smoothing introduced during stripe removal can contaminate subsequent blur kernel estimation, while the strong directional gradients presented by stripe noise can mislead kernel estimation. Moreover, the multi-scale characteristics of stripe noise are rarely explicitly modeled in current methods. To address these limitations, this paper proposes a multi-scale hierarchical joint variational model (MHJVM) for the simultaneous removal of blur and multi-scale stripe noise in wide-field infrared remote sensing images. A Gaussian pyramid is employed to decompose the degraded image into multiple resolution levels. At the low-resolution level, leveraging the weakened structural characteristics of stripes, a multi-scale edge-preserving operator is introduced to adjust the penalty strength, enabling the preliminary suppression of wide stripes and the initial robust estimation of the blur kernel. Second,at high-resolution levels, a blur kernel refinement model based on the anisotropy of stripe residuals is established, explicitly suppressing the contamination of stripe-oriented residuals on kernel estimation and removing fine-scale stripes at the pixel level. Finally, an adaptive ringing suppression model based on Laplacian response is incorporated into the non-blind deconvolution stage to preserve image details while mitigating ringing artifacts. Experimental results on both simulated and real satellite-borne infrared remote sensing images demonstrate that the proposed method outperforms existing sequential frameworks across diverse scenes, including urban, mountain, and sea surface areas. Under multi-scale interference conditions, it achieves an average SNR above 30 dB and an SSIM around 0.92. The method effectively suppresses multi-scale stripe noise, enables accurate blur kernel estimation, and preserves structural fidelity. These results validate the robustness and generalization capability of the proposed multi-scale collaborative optimization framework for joint destriping and deblurring under coupled degradation scenarios.
No takes yet. Share an insight, caveat, or question.
Nie et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: