A Pre-Trained Collaborative Network Leveraging Multi-Scale Depth Information for Image Deraining and Defogging Enhancement
Rain and fog are prevalent weather conditions that negatively impact advanced computer vision tasks such as object detection and recognition. The recently proposed NeRD-Rain model attains state of the art performance in rain removal. However, it fails to comprehensively address the veiling effect in...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10918962/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Rain and fog are prevalent weather conditions that negatively impact advanced computer vision tasks such as object detection and recognition. The recently proposed NeRD-Rain model attains state of the art performance in rain removal. However, it fails to comprehensively address the veiling effect in complex rainy environments. To enhance the NeRD-Rain model, we integrate a pre-trained depth prediction model, grounded in the atmospheric scattering model, into its architecture. This integration is implemented across multiple scales. Experimental evaluations on the RainCityScapes dataset, where rain and fog co-occur, demonstrate that our proposed network substantially improves the dehazing performance of NeRD-Rain under foggy conditions. Quantitative assessments reveal a 1.5% increase in the Structural Similarity Index Measure (SSIM), a 0.01 reduction in Laplacian variance, and a 1.41 decrease in the Frechet Inception Distance (FID). |
|---|---|
| ISSN: | 2169-3536 |