Video contain a lot of information,it have an advantage in information transmission.Therefore,video is widely used in various fields requiring information transmission.The equipment that gets video will be affected by light,weather and other environments when working in different sences.Meanwhile video will be affected by electrical noise in the transmission process.Therefore,the video is easy to affect by various noise,leading to video quality degradation.Noise not only has worse effects on subjective feelings of people,but also has adverse effects on the subsequent processing.Therefore,viedo de-noising is an essential step in video processing.In order to remove the noise in the video.Three aspects of work are mainly carried out.Firstly,an improved non local mean algorithm is proposed for single frame denoising.The improved algorithm focuses on the lack of detail loss due to the smooth with non-local mean algorithm.Based on the prior information of Markov random field,a non local means denoising algorithm based on MRF residual compensation is proposed.The improved algorithm can effectively recover the details lost by the non local mean algorithm in the smoothing process,and improve the PSNR value.Secondly,the fast detection of moving region in video is realized by using the variance of time window.In order to reduce the detection errors caused by holes and tails,the adaptive length window is used to calculate the variance.The algorithm can not only determine the motion region robustly in noisy environment,but also can realize real-time detection of moving region in practical application because the algorithm is simple.Third,a video spatio-temporal bilateral filter based on adaptive weight is proposed.For the motion estimation of 3D bilateral filter using block matching in motion region,the computational complexity is high and the real-time performance is poor.The adaptive weights are designed by the variance of time window.Compared with the traditional bilateral filter,the proposed filter not only effectively improves the PSNR value,retains more details,but also improves the computational speed effectively. |