Image dehazing research based on MB-TaylorFormerV2 model
Abstract
Image dehazing plays a crucial role in enabling all-weather perception and ensuring driving safety in complex weather conditions within intelligent driving systems. In this study, the OTS dataset is dehazed using the MB-TaylorFormerV2 model, which employs an approximate attention mechanism based on Taylor's formula. Compared to traditional atmospheric scattering models analyzed from a physics perspective, significant improvements are observed across multiple metrics. Our obtained PSNR and SSIM are 32.45 dB and 0.96, respectively, representing increases of 77.78% and 14.46%.
Fields of Study & Subjects
Image Enhancement Techniques
Advanced Computing and Algorithms
Image Processing Techniques and Applications
Publication details
- Journal / source
- Second International Conference on Image Processing and Deep Learning (IPDL 2026)
- Publisher
- SPIE
- Pages
- 41–41
- Publication date
- 6 May 2026