FFformer: A Lightweight Feature Filter Transformer for Multi-Degraded Image Enhancement with a Novel Dataset.

Image enhancement in complex scenes is challenging due to the frequent coexistence of multiple degradations caused by adverse weather, imaging hardware, and transmission environments. Existing datasets remain limited to single or weather-specific degradation types, failing to capture real-world complexity. To address this gap, we introduce the Robust Multi-Type Degradation (RMTD) dataset, which synthesizes a wide range of degradations from meteorological, capture, and transmission sources to support model training and evaluation under realistic conditions. Furthermore, the superposition of multiple degradations often results in feature maps dominated by noise, obscuring underlying clean content. To tackle this, we propose the Feature Filter Transformer (FFformer), which includes: (1) a Gaussian-Filtered Self-Attention (GFSA) module that suppresses degradation-related activations by integrating Gaussian filtering into self-attention; and (2) a Feature-Shrinkage Feed-forward Network (FSFN) that applies soft-thresholding to aggressively reduce noise. Additionally, a Feature Enhancement Block (FEB) embedded in skip connections further reinforces clean background features to ensure high-fidelity restoration. Extensive experiments on RMTD and public benchmarks confirm that the proposed dataset and FFformer together bring substantial improvements to the task of complex-scene image enhancement.