A method for removing rain from videos

Abstract

Rain streaks in videos change the intensities of pixels and affect the performance of outdoor vision systems. This paper proposes a rain removal method that can recover videos degraded by rain. This method exploits the fact that all rain streaks in a video have similar directions. First a dictionary is learnt for sparse representation of frame differences. Then the gradient distribution of the intensities of the atoms in this dictionary is analyzed to distinguish rain atoms and non-rain atoms. At last, frames are recovered by only using the non-rain atoms to represent frame differences. Experiments on real videos show the effectiveness of this method. © 2013 Springer-Verlag.