Degraded Data Enhancement Based on Regional Similarity Fusion

The scattering and absorption effect of dust particles on light leads to low contrast and serious color offset of visible light images obtained in sand-dust weather, which affects the reliability of outdoor visual applications such as traffic safety and monitoring systems. Due to complex scene structure and difficult parameter estimation, the existing sand-dust image enhancement methods can not effectively extract the semantic components of the image, resulting in unreal colors and blurred details of the enhanced image. Therefore, we propose a two-stage sand image enhancement method based on regional similarity fusion. Firstly, the gray distribution of the input dust image is compensated to recover the potential information in the scene, and two sub-images with color balance and high contrast are derived. Then, regional similarity calculation, weight allocation, and image fusion are carried out to generate the final clear image based on the regional similarity fusion strategy. The experimental results show that the proposed method can effectively restore the potential features in the dust scene, and the visible edge number ratio (e-score) and edge gradient ratio (r-score) have increased by 0.24 and 0.47 respectively.