Contrastive Degradation-Aware Mamba Network for Blind Image Super-Resolution*

Abstract

Blind image super-resolution (BISR) aims to recover high-resolution images from low-resolution inputs with unknown degradations. Existing degradation representation learning methods usually construct positive pairs from local patches of the same degraded image, which inevitably introduces strong content correlations and limits the modeling capability for complex degradation patterns. To address this issue, we propose a Contrastive Degradation-Aware Mamba Network (CDAM-Net) for blind image super-resolution. Specifically, we design a contrastive degradation representation learning strategy that generates positive pairs from different images with identical degradation parameters, enabling the network to learn content-independent degradation representations. Furthermore, a degradation-aware Mamba restoration network is proposed to adaptively incorporate degradation information into the image reconstruction process. In particular, we introduce a degradation-guided feature modulation mechanism within the Spatial Mamba Module to dynamically guide feature restoration under different degradation conditions. In addition, a gated convolution feed-forward block is utilized to enhance local nonlinear feature representation and texture recovery capability. Extensive experiments on multiple benchmark datasets demonstrate that the proposed method achieves competitive performance against state-of-the-art blind image super-resolution methods in both quantitative and visual comparisons.