Sparse Theory Combine BM3D For Image De-noising Algorithm

Today,digital image processing technology has been widely used,image as an important information is affected by all kinds of the noise in the transmission process,the most common influencing form of noise is Gauss noise and salt and pepper.Image de-noising field had developed various de-noising methods to reduce the adverse information caused by noise,Such as the removing method of Gauss noise with Wiener filtering algorithm and wavelet de-noising algorithm,BM3D algorithm;The removing method of salt and Pepper noise with median filter,PCNN algorithm.But,for the removing method of the mixed noise in image is not well studied,only the above two methods are simple stack processing.In view of the problems,This thesis deeply studies the images of nature,noise existing form,de-noising algorithm theory,it is taking BM3D algorithm as the main body that study on the mixed image noise.The main contents are as follows:(1)Due to the core steps BM3D algorithm applied to the wavelet de-noising algorithm,but also the wavelet de-noising algorithm is a cornerstone in the field of image de-noising.Therefore,this thesis introduces the basic concepts of wavelet transform,mathematical deduction,multi-resolution analysis,and so on,then introducing the wavelet threshold shrinkage de-noising algorithm and wavelet shrinkage de-noising algorithm statistical model.On the threshold method,the selection of threshold value plus image singularity detection,the threshold function is continuous improvement;In the statistical model algorithm,add Bayesian threshold pre-processing,the highest level of MAP estimation process,complex wavelet improvement.Two improved algorithms proposed algorithm in this thesis are better than traditional algorithm in the corresponding PSNR.(2)According to the distribution characteristics of signal singularity on wavelet domain,study on the characteristics of salt and pepper noise in wavelet domain,and then gives the cone wavelet analysis method.through experimental,it's observing the statistical characteristics of salt and pepper on wavelet domain,giving the CIO method of wavelet domain.Next,introducing the detailed steps of the BM3D algorithm,by changing the standard deviation of Gaussian noise,salt and pepper noise density,mixed-noise ratio of the noise.BM3D algorithm to verify the removal of the high intensity of Gaussian noise,impulse noise,mixing noise is ineffective.(3)According to the characteristics of wavelet de-noising algorithm and BM3D algorithm,and the distribution of Gauss noise and salt and pepper noise,proposes a de-noising algorithm for mixed noise.The algorithm is improved based on BM3D,The first is pre-mixed noise image de-noising,traction coordinate strike similar blocks away;Next,threshold for three-dimensional matrix on BM3D algorithm is revised and wavelet coefficients partial is summed;Finally,the estimated coefficients BM3D Wiener filtering algorithm,the size of the image block,the maximum number of similar blocks are revised.Simulation result show that improved BM3D has good effect for mixed noise,and can maintain a good edge detail.