Research On Image Poisson Denoising

As a common carrier of information,image plays an important role in human social activities.However,during the process of image acquisition and transmission,it will inevitably be contaminated by noise.The image contaminated by noise,does not only affect people's visual experience,but also has a negative impact on the subsequent processing of the image;thereby,the effective removal of noise in an image is particularly important.In this paper,we study the problem of Poisson noise removal in images,mainly by discussing the following two approaches.The image denoising method based on Bayesian estimation which takes into account the characteristics of noise and the prior knowledge of the image.It has the characteristics of strong denoising ability and high convergence stability.It is also the widely used denoising algorithm in recent years.Based on Bayesian estimation,this paper constructs a Gaussian prior model of image patch by using self-similarity of image,and a patch-based Poisson denoising model is proposed.The denoising model is a non-quadratic convex function and contains a logarithm term,and it is not easy to optimize directly,therefore,the split Bregman method is used to solve this model,which effectively overcomes the problem of model calculation.Since each patch is not only related to the sample mean of the image,but also to the sample covariance matrix of the image,an optimal estimation of each patch can be obtained.Experimental results show that the proposed algorithm does not only effectively removes the Poisson noise in the image,but also has good visual effect.Based on the existing Gaussian denoising algorithm,this paper discusses the relationship between the Gaussian denoising model and the Poisson denoising model,and proposes a Poisson denoising algorithm combined with Gaussian denoising technique.The Gaussian denoising algorithm is part of the overall denoising algorithm,it only has input and output functions,similar to a "black box".Although the variance-stabilizing transformation method is also combined with the Gaussian denoising algorithm;the basic idea of this,is to transform the Poisson data into Gaussian data with the same variance,and then use the inverse transformation to get the Poisson denoising image.Our proposed method does not need to transform the original Poisson data and the inverse transformation of the variance-stabilizing transformation,but provides a good solution for data distortion caused by inverse transformation.Our experimental results show that the proposed algorithm has better denoising performance at low signal-to-noise ratio compared with variance-stabilizing transformation method.