Optimization Of TV Regularization Denoising Algorithm For Mixed Noise

Total Variation(TV) regularization denoising algorithm can remove the image noise while preserving edges and details of the image. It is the focus of many scholars in the image denoising field. Total Variation regularization algorithm was mainly used to process with single noise model such as Gaussian noise, Poisson noise and so on. In recent years, some scholars began to study the Total Variation regularization algorithm for mixed noise and obtained high-quality denoised image.However, many Total Variation regularization denoising algorithm involves the calculation of partial differential equations and needs complex iterative process while solving the problem especially for mixed noise. They have large amount of calculation and slow convergence. The solving efficiency is not high enough to be used in speed high performance demanding applications. Therefore, this paper focused on the optimization of solving process to speed up the Total Variation regularization algorithm for mixed noise. The main contents are as follows:(1) For the fact that high-resolution images with large amount of data led to large computation in single iteration, this paper introduced an optimization method based on multi-resolution. This method decomposed the solving process into multilple iteration units and got multi-resolution images and edge images from the noisy image by Gaussian pyramid and Laplacian pyramid. The algorithm was solved quickly for low-resolution images in the initial iterations. Its edge information was protected by interpolation algorithm and edge image overlay for high-resolution images in the subsequent iterations. The simulation results showed that the proposed multiscale optimization method could denoise images and the time performance increased more than double, compared to the gradient descent method.(2) In consideration of the solution to the objective function of Total Variation for mixed noise via separate solution of Gaussian noise term and Poisson noise term to improve the convergence rate, this paper introduced an optimization method based on the Split Bregman algorithm. The proposed method achieved the separation of these two terms and fast solution by operator splitting ideas and Bregman iteration. The method also used Shrink operator and Gauss-Seidel method to reduce the computation. The experimental results showed that the proposed method can obtain higher peak signal to noise ratio and rapid convergence in less time, compared to quasi-Newton method and other methods.