A Multi-scale Generalized Shrinkage Threshold Network for Image Blind Deblurring in Remote Sensing

TL;DR

This work tackles blind deblurring in remote sensing by marrying MAP-based optimization with learnable, multi-scale priors in a model-driven unfolding network (MGSTNet). The method alternates kernel and image updates using deep proximal mappings (KPMM and IPMM) and incorporates a multi-scale generalized shrinkage thresholding strategy, guided by learnable transforms ($F(\cdot)$, $G(\cdot)$) and attention. It achieves state-of-the-art restoration across synthetic and real remote-sensing datasets, with robust kernel estimation and superior detail recovery under blur and noise. The framework enhances interpretability and robustness in practical remote sensing deblurring, offering a pathway to extend to related tasks such as super-resolution and dehazing.

Abstract

Remote sensing images are essential for many applications of the earth's sciences, but their quality can usually be degraded due to limitations in sensor technology and complex imaging environments. To address this, various remote sensing image deblurring methods have been developed to restore sharp and high-quality images from degraded observational data. However, most traditional model-based deblurring methods usually require predefined {hand-crafted} prior assumptions, which are difficult to handle in complex applications. On the other hand, deep learning-based deblurring methods are often considered as black boxes, lacking transparency and interpretability. In this work, we propose a new blind deblurring learning framework that utilizes alternating iterations of shrinkage thresholds. This framework involves updating blurring kernels and images, with a theoretical foundation in network design. Additionally, we propose a learnable blur kernel proximal mapping module to improve the accuracy of the blur kernel reconstruction. Furthermore, we propose a deep proximal mapping module in the image domain, which combines a generalized shrinkage threshold with a multi-scale prior feature extraction block. This module also incorporates an attention mechanism to learn adaptively the importance of prior information, improving the flexibility and robustness of prior terms, and avoiding limitations similar to hand-crafted image prior terms. Consequently, we design a novel multi-scale generalized shrinkage threshold network (MGSTNet) that focuses specifically on learning deep geometric prior features to enhance image restoration. Experimental results on real and synthetic remote sensing image datasets demonstrate the superiority of our MGSTNet framework compared to existing deblurring methods.

Figures (7)

• Figure 1: The overall framework of the proposed MGSTNet that learns enriched feature representations for image blind deblurring. MGSTNet is based on an iterative design with alternating image/blur kernel optimizations, and its main idea is to learn the representation mapping between the image space (or kernel space) and the feature space by using an encoder-decoder module and a shrinkage threshold module.
• Figure 2: Convergence analysis of MGSTNet for image deblurring with different structure.
• Figure 3: Visual comparison of different deblurring methods on AIRS datasets.
• Figure 4: Visual comparison of different deblurring methods on AIRS datasets with small perturbation noise.
• Figure 5: The PSNR box-plot of different deblurring methods on AIRS. '(+)' represents the deblurring performance experiment with noise.