Research Of The Image Denoising And Restoration Based On High-order Variation Method

Variation method, which has good theoretical basis and high degree of flexibility, hasbeen widely used in image processing areas, especially in image denoising, imagerestoration, image segmentation and so on. The main research of this paper is highvariation image processing method, including image denoising; the point spread function(PSF) parameter estimation and image restoration, etc. The main work of this paper is asfollows:Firstly, a new adaptive high-order variational image denoising model based on thecharacteristics of human visual system (HVS) is proposed by introducing control parameterω.According to values of M (x, y), which is the noise shielding function, this model adaptiveselects parameter ranges, and then selects the integrated model of diffusion constraints items.Experimental results shows that: the model can inhibit the "ladder effect" generates, restorationimage quality improvement and meet the human visual system.Secondly, in order to identify the parameters of the PSF accurately, an identificationmethod based on cepstrum domain is presented in this paper. It uses Radon transforms andwavelet packet decomposition to analyse fuzzy image cepstrum, in order to extract fuzzymeasure and fuzzy angle. Then PSF parameter estimation experiments are carried out on thelarge numbers of computer simulation images, the results shows that: this method accuratelyestimats the parameters of the PSF, and possesses a certain degree of noise immunity.Thirdly, a high-order variational motion blurred image restoration model based on theestimated PSF parameter is proposed. Compared with the classical image restorationalgorithm, good recovery results are obtained. The restored images also validate theusability and accuracy of the proposed method. Fourthly, a high-order hybrid regularization method for image blind restoration isproposed based on the insufficient information of PSF. Because of the sparse edges inimages, the total variation regularization restriction is applied to the edge texturecomponent, and the high-order regularization restriction is used to the smooth regions ofthe image. The results prove that the proposed method is able to preserve the image edgeswhile avoiding staircase effects and false edges in the smooth regions.Finally, based on the characteristics of the high-order variation model, suitablenumerical method is put forward to the optimal solution of the model.