Combined Motion Estimation And Wavelet Domain Video Denoising

With the popularization of digital electronic equipment, multimedia informationhas become more important. However, videos are often corrupted by noise duringacquisition, degrading the user’s perceived quality of the video. To address this issue,we propose a novel video denoising algorithm in this thesis. The first step of ouralgorithm is time domain filtering, using the improved Adaptive Rood Pattern Searchalgorithm for motion estimation. At positions where no motion was detected, astandard recursive temporal filter is applied. At moving positions we filter along theestimated motion trajectory. The second step is spatial filtering based on theframework of wavelet analysis and non local means, every frame of the video isdecomposed into low-frequency and high-frequency. In the smooth low frequencypart, use the strong non local means filter to denoise better. While denoise in lowdegree in the texture information area by wavelet threshold method.Experiments demonstrate that the proposed denoising scheme can avoid motionblurring, overcome the problem of false texture information produced in the flat areaand protect the texture information much better. The proposed method has greatlyimprovement in subjective feeling as well as in the objective quality. What’s more, theefficiency of the proposed algorithm increase significantly, it has strong real-timeproperty.