摘要：Nowadays, multimedia, especially video, is used widely in many domains. But camera lens are easily stained because they are always outside, such as in traffic and monitoring. Admittedly, the video from these cameras is out of action because it cannot provide clear frames. In this paper, we provide a detecting method from video with background restructure. This method is presented to detect stains in video. Firstly, the first video frame is used as a pseudo-background because original cameras are clear with no stains. Then, background is constructed by the pseudo-background and the continuous frames which are from the video. Moreover, with training in the frames, we restructure background by dropped moving objects in it. Finally, after background restructured, we detect stains in camera lenses by the restructured background, and provide positions of stained area. Experimental results show its robustness and practicability
