SSDC-GAN: Same Size Densely Connected GAN for Dehazing Network
Juan Wang, Chang Ding, Yonggang Ye, Minghu Wu, Zetao Zhang, Sheng Wang, Hao Yang, Ye Cao · Atlantis Highlights in Intelligent Systems/Atlantis highlights in intelligent systems · 2023
Haze weather negatively impacts the quality of external image collection and requires prompt resolution.However, most current deep learning image dehazing models struggle with restoring detail and color accuracy in real-world hazy images, hindering their practical application for high-quality image projects.To overcome this issue, we propose a novel connected mode (SSDC) for endto-end dehazing that simplifies the problem to an image conversion task without relying on atmospheric scattering models or precise priors.The SSDC-GAN generator employs an encoder-decoder, same size densely connected architecture with residual blocks, and a depth discriminator to balance the relationship during training.Experimental results demonstrate that the proposed method performs favorably against state-of-the-art dehazing approaches on various benchmarks using real-world datasets O-HAZE and I-HAZE while preserving accurate contour and color information.