Video Denoising Algorithm Based On Surfacelet Transform

Video images signal is often affected by noise in the processing of sending, transmission and reception in video so that video images become blurred and the parts of the important information cannot be identified. These noises would affect the visual effect and the further processing of the video sequence. Therefore, the video sequence denoising has always attracted our attention, and people have exploited the subject a few years. How to preserve important information and how to capture the surface singularity in video images that have been a hot point and difficult problem.Surfacelet transform(ST) is a new3D transformation and it is proposed through the combination of the multiscale pyramid and the N-dimensional directional filter banks. So surfacelet transform can effectively capture and represent the singularity on the smooth surface, which has better features, such as multidirectional decomposition, efficient tree structural filter banks, perfect reconstruction, and low redundancy, and is an ideal method for video images processing. Because of the analysis of the characteristics of video images, this paper discusses the methods of threshold processing and coefficient shrinking of video denoising in surfacelet transform domain。1. A video denoising based on adaptive threshold considering correlation of surfacelet neighborhood coefficients is proposed. According to the correlation of neighborhood coefficients and the distribution characteristics of noise at different scales in surfacelet transform domain, and the space information and direction structure information in the neighborhood is selectively added. So the threshold corresponding to each coefficient can be accurately calculated. Therefore, these noises in video can be more effectively removed.2. A video denoising method based on adaptively shrinking surfacelet transform coefficients is proposed. Firstly, the threshold and the mask classification of each coefficient can be calculated. The threshold is applied to construct energy ratio of each coefficient. The mask classification is used to construct prior ratio of each coefficient. Based on the construction of three-dimensional and directional neighborhood, the ideal neighborhood shape is selected, so that optimal prior ratio is obtained. Secondly, the energy ratio and prior ratio are combined and applied into shrinkage estimator, which can be used to shrink the coefficients. Finally, the denoised video is achieved by inverse surfacelet transform using shrunk coefficients. 3. A video denoising method based on logarithmic normal distribution model in surfacelet transform domain is proposed. Firstly, the probability distribution feature of noise in surfacelet transform domain is evaluated through Monte Carlo method, based on which the threshold of each coefficient can be calculated so that the mask classification can be obtained. Secondly, the prior ratio and likelihood ratio of each coefficient can be calculated. Because the logarithmic normal distribution probability density function can better describe the distribution of the ST coefficients. The logarithmic normal distribution model is used for fitting the likelihood ratio probability of the ST coefficients. The energy ratio and prior ratio are combined and applied into shrinkage estimator, which can be used to the shrinkage of the coefficients.