The Research Of Video Sequence Denoising In Surfacelet Transform Domain

Video denosing is an important work in the field of image processing. As a newanalytical method, Surfacelet transform (ST) provides a new and powerful method for thevideo information processing, which can captures and indicates the singularity signal ofsmooth surface effectively. Besides, it has the character of multi-directionaldecomposition, anisotropy, the highly efficient filter bank of tree structure, completereconstruction, low redundancy and so on, so it is a very ideal method for videoprocessing. The paper studis the characteristics of video signal in the Surfacelet transformdomain and proposes the following algorithms:1. A video denoising method based on fuzzy feature in Surfacelet transformdomain is proposed. The proposed method fully considers scale and spatial correlationneighboring each point at different layers and different direction subband. The details ofthe neighborhood are taken full account, too. The method is able to reserve the videodetail information and edge, and it can effectively remove the noise and obtain the bettervideo image.2. A video denoising method based on non-parametric estimation in Surfacelettransform domain is proposed. First, the mask classification of each coefficient can becalculated; then we calculate the likelihood ratio based on non-parametric statisticalmodel and priori ratio of each subband coefficient, then calculate the shrinkage factor ofST coefficients; finally, the ST coefficients are shrunk.3. A video denoising method based on Laplace model in Surfacelet transformdomain is proposed. We make two improvements to the classic subband adaptive Laplacemodel algorithm. First we use the edge standard deviation of neighborhood window toreplace the edge standard of sub-band to obtain the local self-adaptive threshold; Second,by researching the statistical analysis of Surfacelet transform coefficients, we improvethe threshold parameter, so that the method can obtain a ideal result.