Entropy-based rain detection and removal

Presence of rain causes distortion in videos which affects the performance of various outdoor vision systems. This paper presents a rain detection algorithm based on local grayscale entropy and temporal intensity range of pixels. Local grayscale entropy can sufficiently detect all rain pixels using only one frame if the background is static and using 5 frames if background is dynamic. The gray scale entropy and temporal intensity range values are combined to differentiate rain pixels from other objects of motion in video. For rain removal, the detected rain pixels can be replaced by the corresponding temporal mean of the intensity values of the 15 consecutive frames. Experimental results show that the proposed algorithm performs significantly better than the existing ones both qualitatively and quantitatively.