RASP-Net: Residual Block with High Similarity Pass Attention for Image Desnowing

Key Points

• The aim is to improve image quality by developing a novel desnowing algorithm utilizing attention mechanisms.
• Developed RASP-Net integrating residual blocks and spatial attention mechanisms to enhance feature extraction.
• Employed high similarity pass attention to dynamically assess and refine snowflake features.
• Conducted extensive experiments on Snow100K, SRRS, and CSD datasets to evaluate performance.
• Achieved a peak signal-to-noise ratio of 35.51 dB on the Snow100K dataset, outperforming comparison methods.
• Demonstrated effective detail reconstruction obscured by snowflakes in both synthetic and real-world datasets.