Image De-Weathering for Road Based on Physical Model

Images of roads captured in bad weather suffer from poor contrasts and colors. It is known that the degradation of an image due to scattering of light is exponential in the depths of the scene points. Therefore, to estimate the depths, previous algorithms using physical model develop an interactive step. But the method is limited, because it needs to estimate four or more unknown parameters and select the vanishing point manually. In this paper, we present a simple implementation based on Hough transform and maximum entropy theory to solve the problems. First, the physical model is simplified to reduce the number of unknown parameters to one. Then, an automatic method based on Hough transform is proposed to select the vanishing point. Finally, the maximum entropy theory is used to acquire the optimal parameter. The experimental results on real images show the effectiveness of the approach.