Research On Model And Algorithm For Restoration Of Image With Speckle Noise

Image has become one of the important ways to obtain information,but noise pollution and visual blur are often appear in the process of image acquisition and transmission that make the image quality degradation.As one of the core work of digital image processing technology,image restoration is the process of reconstructing the degraded image by computer.Image restoration on speckle noise has attracted wide attention in the field of applied mathematics and image processing.Speckle noises often appear in medical images.Along with advances in medical technology,more and more people research subject about the medical images.This dissertation focus on the problem of image restoration based on speckle noise,and different the previous literature construct the regularization term of the model by using the total variation method.Firstly,the dissertation introduces two kinds of common noise,namely,additive noise and multiplication noise,and the reason and background of noise model.Secondly,in the dissertation,we propose a new model based on the framework regularization constraint to deal with speckle noise.In order to solve the new model,we investigate two new approximation schemes that are derived from the variable-splitting and penalty techniques in optimization.The optimization problem of two variable approximation models can be obtained by using the alternating minimum algorithm.Specifically,we transform the approximate model into two sub-models,and the two sub-models optimization problem will be solved by Newton iterative algorithm and adjacent algorithm respectively.In addition,we theoretically analyze the convergence of the two variation model algorithm,the existence and uniqueness of the minimum solution of the model and the strict convexity of the objective function.Finally,a large number of numerical experiments of simulated real images and medical images show that the propose algorithm is effectiveness and feasibility in image restoration,and have better recovery performance compared with other algorithms.