Multi-path dilated convolution network for haze and glow removal in nighttime images

In this paper, we address the single-image haze removal problem in nighttime scenes. The night is a severely ill-posed due to presence of various visible light sources with varying colors and non-uniform illumination. These are different shapes introduce noticeable amount glow To overcome these effects, deep learning-based DeGlow–DeHaze iterative model which accounts for glows. proposed linear combination three terms: direct transmission attenuation, airlight glow. First, multi-path dilated convolution DeGlow network introduced interactively learn local features reception fields reduce effect. term estimated by binary mask that informs location illumination source. As result, image only left terms. Finally, separate post-processing DeHaze included remove effect from reduced image. For our training, collected hazy images internal external resources, synthesized maps NYU depth datasets, consequently restored haze-free images. quantitative qualitative evaluations show effectiveness on several real synthetic compare results existing models. experimental demonstrate CNN outperforms other state-of-the-art methods terms PSNR (19.25 dB), SSIM (0.9958) evaluation parameters computation time.