Study On Image Restoration And Enhancement Based On Variation

The variation and partial differential equation(PDE) has achieved perfect effectin image processing, which has good local self-adaptive and high degree flexibility. Inthis paper, we describe image texture features take advantage of variation, PDE andother methods to achieve image restoration and enhancement. The details and theinnovation are as follows:(1) we study and analysis the staircase and edge fuzzy of the existing models, whichuse image gradient as the only features. In order to better describe image detailfeatures, we introduce image structure tensor into variation model. So the diffusionprocess does not depend on gradient information simply using image structure tensorto control the diffusion, therefore, the model is able to describe the imageedges,textures and smooth area accurately and subtly.(2) To overcome PDE method in image restoration process is susceptible to noise, theperform of time-frequency localization features is poor; wavelet transform method iseasy to lose high-frequency information cause ringing, pseudo-Gibbs visual distortionand image drift edge problem. In this paper, we use total variational to process imageto maintain the position and shape of the image, combined with the spatial lowcorrelation of wavelet multiscale decomposition coefficient and a goodtime-frequency localization features proposed a new wavelet domain imagerestoration model based on variation regularization. In order to better balance theregular item and fidelity item, we take advantage of wavelet coefficients module valueto construct an adaptive weighting function which can better balance regular item andfidelity item under different scales and different sub-frequency.(3) we discuss the strengths and weakness types of image enhancement methodsbased on PDE: based on historgram equalization, based on shock filter process andbased on forward-and-backward. We use the wavelet superiority of description imagetexture, combine with forward-and-backward method, propose a wavelet domainforward-and-backward image enhancement model.