A Research Of Models And Algorithms For Multiplicative Noise And Blur Removal

As the basis of visual perception,image is an important part of information.In order to improve the image quality,image processing technology has emerged,maintains the momentum of rapid development,expands the scope of applications increasingly.Image restoration is the basis operation of image processing.Image is inevitably distorted by noise and fuzzy interference in the process of acquisition and transmission,leading to the decrease of the image quality(i.e.,image degradation).Image restoration is to restore the true high quality image from the observed degraded image.According to the task of image restoration,this thesis focuses on the image denoising and deblurring issue.As multiplicative noise is a challenging subject with high practicability,this thesis studies the model and algorithm for multiplicative noise and blur removal.This thesis firstly outlines the main research progress in the field of multiplicative noise and blur removal.Then several typical methods in recent years on multiplicative noise and blur removal are introduced in detail.Those studies include multiplicative noise removal method using variable splitting and constrained optimization,multiplicative noise and blur removal based on convex optimization and alternating direction method of multipliers,two-step approach,multiplicative noise removal model based on adaptive learning dictionary and total variation regularization,a variational model on multiplicative noise removal using multi-grid algorithm.Works have been shown in two aspects,which include modeling foundation and numerical algorithm.This thesis proposes a new high-order total variation regularization based convex model for multiplicative noise and blur removal,which can be used for many kinds of multiplicative noise and blur.By using a mixture of the high-order total variation regularization term and the total variation regularization term,the model can preserve the image edge information while avoiding staircase effects in the smooth regions of the restored image.The numerical method for solving the new model is the alternating direction method of multipliers.A large number of numerical experiments show that the new model generally achieves better image restoration performance compared with the contrast model,which can be observed from both the data evaluation indexes and visual perception.