Research On Non - Local Image Denoising Algorithm Based On Improved

Image denoising is an important research content in the field of image processing. The researchers had put forward many effective methods to solving image denoising problem according to the different kinds of noise. Low-rank modeling theory which extended by compressed sensing has made a great progress, and shows a great advantage in machine vision and task learning. In this paper, we made a research on the restoring the real image by removing the noise signals and maintaining the contour and texture details of the image based on the algorithm model and optimization algorithm research. The main works can be summarized as follows:1. We analysis and compare the performance of existing image denoising methods.Image denoising method includes local image denoising method and non-local image denoising method. Local image denosing algorithm is simple and has low computational complexity, but the effect of it is not good, because of the denosied image is always too smooth and loss too much texture information. Non-local image denoising method can get better performance than conventional local denoising method in both keeping the contour and texture. Even if the image is seriously polluted by noise, non-local image denoising method also can obtain good denoising effect.2. This paper proposes a non-local image denoising algorithm based on iterative log threshold algorithm. In the process of solving the RPCA model, the quality of the threshold algorithm has great effect on signal recovery performance. Hard threshold algorithm is neither continuous nor differentiable on the threshold, and would easily produce the oscillation signal which leads to worse smoothness in the signal reconstruction method. Compared with the hard threshold algorithm, the reconstruction performance of the soft threshold algorithm is better, but the soft threshold algorithm has some negative effect due to without using the best value when revising the coefficient. Considering the disadvantages of soft and hard threshold algorithm, we introduce the concept of log threshold, which inherits the advantages of soft threshold function and hard threshold function. The log threshold function makes transition that between the soft and hard threshold be smooth. During the log threshold function, we get a new solving method. Solving the RPCA model by the log threshold algorithm, the experiment results show us that our log threshold algorithm is more effective in image denoising than the conventional methods.3. We put forward the weighted robust principal component analysis(WRPCA) based on the non-local image denoising algorithm. Firstly, we analysis of the robust principal component analysis(RPCA) model, then add the weighted nuclear norm to robust principal component analysis model and construct the weighted RPCA model based on the carefully studying nuclear norm and combining it with the self-similarity characteristics of natural image. We can solve the problem by Augmented Lagrangian Multiplier method. In the process of image denoising, first of all, we divide the noise image into blocks, then cluster the image blocks by block matching method, similarity blocks matrix will be obtained simultaneously. After that, we use WRPCA for similarity blocks matrix to recovery the low rank matrix. Compared with the existing classic algorithms on both high noise images and low noise images, our method shows better denoising effect, is more robust. Meanwhile,WRPCA have better performance on preserve both image structure and image texture details.