Based On Sparse Constraint Regularization Algorithm For Image Restoration

In daily life and scientific research, the device itself or environmental factors oftenlead to the images with different degree of fuzzy. The fuzzy result in degeneration of thequality of images, and the following image processing will be affected by it. Imagerestoration technology takes advantage of the information of blurred image, the prioriinformation on the clear image, and the feature of the image processing system, to buildthe model, so as to achieve the purpose of restoring the clear original image from thefuzzy image. In the paper, on the basis of the feature of the image, the research of theimage restoration algorithm in the following three aspects:Firstly, an image restoration algorithm based on wiener filtering and wavelettransform is studied. In this algorithm, the colored noise generate from the deconvolutionprocess can be effectively expressed by Fourier transform, at the same time, the signal andimage can be reasonably expressed by wavelet domain. The Fourier shrinkage and waveletshrinkage can be used to estimate original image, which can obtain the balance of theFourier transform and wavelet transform, and also gets the optimal recovered image.Secondly, on the one hand, a blurred image restoration algorithm based on totalvariation is learned, which the gradient projection method is used to optimize blurredimage denoising, according to the distribution of image and blurred kernel are sparse inthe gradient domain; and the monotonous fast iterative shrinkage threshold algorithm isused to solve the problem of image deblurring. On the other hand, a image restorationmethod which used edge detection is researched, according to the basic characteristics ofnatural images that the image mainly boundary is sharp and sparse, at the same time, thesparse degree of blurred image is poor than clear. The image edge detection in the firstplace, in order to get a regular item, and then restore image. By the analysis ofexperimental result, we can see that the image restoration algorithm based on totalvariation have good performance than the method used a single regularization.Finally, a blind image deblurr algorithm that used three kinds of sparse regularizationis researched. In order to get a clearer image edge and detail, the blurred image should be sharpened in test. In the experiment, l1/l2regularization is used as penalty terms of highfrequency domain, at the same time, sl0Regularization and l1regularization are used in theestimate of blur kernel. The method of multi-scale employed in test. The experimentalresults show that the algorithm which combines with three regularition have betterperformance than which only used two kinds of regularition.