A Variational PDE Image Denoising Model With Structure Vector

As an extremely important transmission medium,digital image has become an indispensable source of information in the information age.At the same time,with the continuous improvement of the ability of computer processing and the increasing requirement of information society for image processing,digital image processing has also become a quite active research field.However,due to the limitations of the imaging methods and various modern conditions and the existence of external interference,the actual image is more or less contaminated by the noise signal,which impede the interpretation of the image information.,At the same time,it has a great influence on the subsequent image processing,the research of noise reduction algorithm for digital image processing is of great theoretical and practical significance for improving the overall performance of image processing theory system.And it can ensure that people understand the information contained in the image correctly and obtain the effective information in the image accurately.Most of the traditional methods of noise reduction are likely to cause blurred edges of images,so it is hard to maintain too much detail information of images.In recent years,the image denoising algorithm used in the variational PDE is based on the gradient and curvature of the image as an operator.In the edge region and the flat region,the size of the gradient or curvature is automatically selected by the full variational model(Total Variation,TV)or by the harmonic model,so that the image edge detection process and map are made.The process of removing dryness,such as the full variation denoising model,the generalized denoising model and the adaptive TV denoising model,is a common processing model.Because the only edge detection in this model is based on the gradient or curvature,there are two disadvantages:first,it can not distinguish the image.The prime point is the edge of the image or the single noise with larger gradient value in the flat area.Two,there is no effective way to identify the pixels to be processed.Therefore,even if the selection of the later model is very reasonable,but the edge detection operator can not accurately select the denoising model,which leads to the phenomenon of step effect and edge blurring in the image denoising process,and can not achieve more ideal effect.In order to solve these problems,the paper takes the source of the problem as its starting point,taking the structural information provided by the eigenvalue of the structure tensor as its edge detection operator,This edge detection operator can effectively overcome those shortcomings,and its diffusion term can be weighted by the image fine information,developing a new model which based on structure tensor,improving the ability of suppressing the ladder effect and maintaining the detail information in the total variation denoising algorithm.In numerical experiments,this paper compared several representative algorithms,making the subjective and objective evaluation from the angle of visual effect,Peak Signal to Noise Ratio(PSNR)and Mean Absolute Error(MAE).Numerical experiments demonstrate that the new model performs well in the aspect of restraining the ladder effect and preserving fine details.