Automatic image de-weathering using physical model and maximum entropy
Xin Wang, Zhenmin Tang · 2008
Images captured under bad weather conditions usually have poor contrasts and colors. Due to the scattering of light, the degradation of an image increases exponentially with the depths of the scene points. Previously implemented methods are limited, because an interactive step was required to select the sky brightness and the vanishing point of the image, as well as the information about the atmospheric conditions. In this paper, we propose an automatic method based on physical model and maximum entropy to remove weather effects using only a single image. First, we segment the sky region by optimal estimated normal distribution and select the lowest point of the sky region as the vanishing point. Then, we exploit the physics-based model to remove weather effects from the image. At last, to overcome the defect of a single image lacking exact atmospheric information, we propose an algorithm based on maximum entropy to select the optimal scattering coefficient of the atmosphere. Our automatic method for image de-weathering is suitable not only for gray level images but also for RGB color images. Compared with other methods, our method is robust and has good efficiency.