Restoration of Adverse Weather Images Based on Dynamic Group Correction Transformer

Adverse weather conditions degrade image quality, negatively impacting high-level vision tasks. To address the limitations of existing methods in handling adverse weather degradation, we propose a Transformer-based architecture incorporating Grouped Correction Histogram Attention to enhance degraded image restoration. The proposed method leverages histogram sorting to improve feature modeling capability, employs a grouped correction strategy to suppress redundant information, and introduces bias correction to stabilize attention computation. Additionally, we integrate a channel-gated mechanism into the feedforward network to fully exploit cross-channel feature complementarity. During optimization, we design a hierarchical loss framework, to enhance the perceptual quality of restored images. Experimental results on multiple public datasets demonstrate that the proposed method outperforms SOTA approaches in terms of PSNR and SSIM, achieving higher-quality image restoration under adverse weather conditions.