The Digital Image Restoration Technology Based On Variational Theory

The problem of degraded image restoration is a theoretical and practical theory. There are lots of image restoration algorithms to restore the degraded images. There are two issues in image restoration, which are denoising image and defuzzifying image. In the process of image restoration, the image denoising and the image details preservation are often opponent procedures. It is hot topics to preserve image details while restore the degraded image. In this paper, the two kinds of image restoration techniques will be introduced respectively and improved algorithm will be proposed.In the part of the image denoising, the advantages and disadvantages of the isotropic diffusion model will be compared with TV denoising model and the adaptive TV denoising model. Through studying the varying regularization parameter and summarizing the disadvantages of the above model and then combining with the advantages of removing the isolated noise of median filtering model, propose a new adaptive total variational median filter model which is based on gradient information. In the new model, noise is removed while the image edge information is preserved. At the same time, the possibility of generating false edges is reduced and higher quality images are obtained.In the part of image defuzzification, at the situation of the point spread function is known, research the processing results of the TV model and obtain the good experimental results. When the point spread function is unknown or part of the point spread function is unknown, the restoration problem is named blind image restoration. When using the blind image restoration, the TV model with a dual regularization parameter will be used, which guarantees the robustness of the image during the recovery process. Firstly, according to the priori knowledge and the nature of the point spread function, we can obtain an estimate value of the point spread function and then repair the support domains of the point spread function. So the blind image restoration problem is converted into a known blur kernel image restoration problem. Because of repairing the support domain of the point spread function. makes it approaching to the true value and improves the quality of the restored image.