Enhanced Feature Joint Optimization Network for Real-World Blind Super-Resolution

Real-world image degradation is often caused by a complex combination of factors, including blur, noise, camera imaging system limitations, image processing, and image compression. This results in an ill-posed problem, making it difficult to obtain satisfactory solutions by relying solely on mathematical formulas. To address these complex and unknown degradation issues, we propose a novel enhanced feature joint optimization network that incorporates a newly added Feature Adaptive Aggregator (FAA) and an Enhanced Spatial Attention mechanism (ESA) for real-world blind super-resolution. The key to our approach is the simultaneous optimization of degradation estimation and image restoration models, allowing them to influence each other and better adapt to real-world image degradation. This integrated optimization can avoid the mismatch issues associated with independent optimization. Final experiments show that our method improves both qualitative and quantitative performance.