Visibility restoration for real-world hazy images via improved physical model and Gaussian total variation

1 Introduction Under real-world haze conditions,the existence of haze particles in the atmosphere reduces the visibility of captured image.Furthermore,the noise is inevitably introduced into the degraded image,which further deteriorates the visual quality of the images.To enhance the visibility and quality of outdoor real-world hazy images,numerous algorithms have been proposed to remove haze from a single input image.The existing methods are broadly lumped into two categories:prior-based methods[1,2]and learning-based methods[3–6].Unfortunately,the widely used atmospheric scattering model and the corresponding haze removal methods fail to take the noise interference into account,which may result in poor visibility restoration performance.