Sequential dual attention network for rain streak removal in a single image

Rain streaks in images captured under adverse weather conditions, such as rain, haze, or snow, can significantly degrade visual quality and negatively impact performance in outdoor visual surveillance and other applications. This paper proposes a novel framework called SSDRNet (Sequential dual attention-based Single image DeRaining deep Network) for single-image rain streak removal. The approach is based on the principle that the correlation among rain streaks within an image is stronger than that between rain streaks and background pixels.

Research Focus

A two-stage learning strategy to better capture the distribution of rain streaks within a rainy image
Three specialized components:
- Residual dense blocks (RDBs) for feature extraction
- Sequential dual attention block (SDAB) for focusing on rain streak patterns
- Multi-scale feature aggregation module (MAM) for aggregating multi-scale feature