Deep learning-based blind image super-resolution with iterative kernel reconstruction and noise estimation

TL;DR

This work tackles blind SISR under unknown and complex degradations by proposing IKR-Net, an iterative, end-to-end framework that jointly estimates the blur kernel, noise level, and high-resolution image. The architecture combines model-based degradation steps (kernel and SR reconstruction via fixed operators) with learning-based denoisers, organized into four modules: Kernel Initializer, SR Reconstruction, Kernel Reconstruction, and Noise Estimator. Key contributions include a modular HQS-inspired iterative scheme, a kernel initializer for a good starting point, an iterative kernel refinement path, a noise estimation module that adapts hyper-parameters, and a ResUNet-based denoiser for high-quality SR. The approach achieves state-of-the-art results for blind SR, particularly on noisy inputs with motion blur, and demonstrates strong generalization across isotropic/anisotropic Gaussian and motion kernels, offering practical implications for real-world image restoration and potential extension to other inverse problems. $y = (x \otimes k) \downarrow_{s} + n$ serves as the core degradation model around which the method is built, enabling robust restoration without prior knowledge of the blur kernel or noise level.

Abstract

Blind single image super-resolution (SISR) is a challenging task in image processing due to the ill-posed nature of the inverse problem. Complex degradations present in real life images make it difficult to solve this problem using naïve deep learning approaches, where models are often trained on synthetically generated image pairs. Most of the effort so far has been focused on solving the inverse problem under some constraints, such as for a limited space of blur kernels and/or assuming noise-free input images. Yet, there is a gap in the literature to provide a well-generalized deep learning-based solution that performs well on images with unknown and highly complex degradations. In this paper, we propose IKR-Net (Iterative Kernel Reconstruction Network) for blind SISR. In the proposed approach, kernel and noise estimation and high-resolution image reconstruction are carried out iteratively using dedicated deep models. The iterative refinement provides significant improvement in both the reconstructed image and the estimated blur kernel even for noisy inputs. IKR-Net provides a generalized solution that can handle any type of blur and level of noise in the input low-resolution image. IKR-Net achieves state-of-the-art results in blind SISR, especially for noisy images with motion blur.

Figures (13)

• Figure 1: Iterative estimation of SR image and blur kernel in IKR-Net architecture.
• Figure 2: The overall iterative architecture of IKR-Net model.
• Figure 3: Kernel initializer $\mathcal{I}$ takes LR image as input and generates an initial guess $k_0$ for the kernel.
• Figure 4: ResUNet architecture for SR image denoising module $\mathcal{P}$.
• Figure 5: Noise and hyper-paramater estimation module $\mathcal{F}$.