Video Denoising Based On Non-local Means

With the development and application of the computer network communication and multimedia technology, the videos that people contacted are more frequently. But the quality and clarity of the video have influence on the effect. Because of the electronic production, the noise can produced both in the making and transporting of the video. The effect is decreased by the noise and is not good for the subsequent processing. Therefore, it is very important that how to denoise effectively and improve the quality and clarity of the video. The denoising in the spatial domain and temporal domain is mainly studied in this thesis.Firstly, the origin of the noise and the classification and characteristics of the noise are expounded and the denoising in the spatial domain is mainly studied. Non-local means is fine performance in spatial domain, but it produces fictitious texture information. So a neighborhood adaptive algorithm based on Non-local means is proposed. In order to denoise better, we denoise in high degree in the flat area. And in order to protect the texture information, we denoise in low degree in the texture information area. So we solve the texture information produced in the flat area but in the texture information area, we can protect the texture information much better. The experiments show that both subjective quality and objective quality are greater improvement of video with the greater texture information especially.Secondly, video is different from single image. Video is consisted of some frames correlative in temporal domain between the frames. So it is very important that we can denoise in the temporal domain. It can decrease the different of adjacent frame after denoising, so it does not flickering when we watch it. Generally, the motion compensation is base on the block, so it can produce the blocking effect. So the motion compensation filter point by point in the spatial and temporal domain is proposed. We use idea of the Non-local means. So we do not choice the optimal one, every pixel has a weight. We can denoise both in the current frame and in the adjacent frame by the weighted average. The experiments show that both subjective quality and objective quality are greater improvement and decrease flickering when playing the video.Finally, we summarize the thesis, and point out future work.