Restoration Of Image Degraded By Rain Based On Sequence Images

Rain produces sharp intensity fluctuations in images or videos which severely degrade the performance of outdoor vision systems. Removing the visual effects of rain is important in order to make outdoor vision robust to rain. The achievement will be more helpful for tracking, recognition and navigation. For bad weather, rain is one of very complex weather conditions. It is a collection of randomly distributed water droplets of different shapes and sizes that move at high velocities, and the precipitation is randomly time-varying. A group of such falling rain drops creates a complex time varying signal in images and videos. And the appearance of each drop is dependent on the illumination of the environment. Hence, the visual effects of rain are complex.In the paper, we handle rain in videos from both special and frequency aspect. Based on the corresponding physical model, it re-implements respective algorithms for detecting and removing all rain from videos, and get clear images. It is composed of several aspects as follows:(1) Based on comprehensive analysis about the physical property, the spatio-temporal property and the chromatic property of rain, it develops special physical model which includes the appearance model, special-temporal distribution and the chromatic model, then applys K-means for detecting and removing rain from videos combining linear constrains. The experimental results state that the method can handle most of steady rain videos.(2) Combining the visual model of rain streak and their statistical property, it handles rain based on the global models in special and frequency space. Firstly, using improved background subtraction we get rough detection, and then by the difference of 3D Fourier between rain and non-rain it gets finer detection. Finally, using the background model, it removes most of rain from videos with iterative optimization methods.