Search for a command to run...
High-quality images are essential for human-computer interaction in industrial systems; however, captured images are often degraded by device vibrations and moving objects. The single-image super-resolution (SISR) task aims to reconstruct high-quality images from low-quality inputs, where deep networks have demonstrated significant success. Nevertheless, noisy image super-resolution remains challenging due to the difficulty of separating noise from genuine structural details. Unlike traditional methods that struggle to distinguish noise from actual image details, the proposed unified framework for noisy image super-resolution (UFNet) combines discriminative learning with a degradation model to enhance both noise suppression and detail recovery. UFNet employs two parallel networks to extract informative features for reconstructing high-quality images. The upper branch utilizes a discriminative learning strategy to remove noise, while the lower branch applies the concept of a degradation model to recover structural details. To restore lost details while maintaining naturalness and structural consistency, a Feature Distillation and Refinement Block (FDRB) is embedded in the lower network. Furthermore, a refinement network is employed to eliminate redundant information introduced during the fusion operation, thereby unifying the results of the two branches and enhancing the final super-resolution outcomes. Extensive experiments demonstrate that the proposed UFNet achieves excellent performance in noisy image super-resolution. The source code of UFNet is available at https://github.com/WuZiang73/UFNet .