基于多元统计模型的分形小波自适应图像去噪  

Adaptive Fractal-Wavelet Image Denoising Base on Multivariate Statistical Model

作　　者：王智文[1,2,3] 李绍滋[1,3] 

机构地区：[1]厦门大学智能科学与技术系,福建厦门361005 [2]广西科技大学计算机学院,广西柳州545006 [3]福建省仿脑智能系统重点实验室(厦门大学),福建厦门361005

出　　处：《计算机学报》2014年第6期1380-1389,共10页Chinese Journal of Computers

基　　金：国家自然科学基金项目(60873179);广西壮族自治区自然科学基金(2011GXNSFA018162;2013GXNSFAA019336);高等学校博士学科点专项科研基金(20090121110032);深圳市科技计划项目-基础研究(JC200903180630A)资助~~

摘　　要：为了克服分形小波去噪方法缺乏保护图像的边缘与细节的缺陷,文中提出了一种基于多元统计模型的分形小波自适应图像去噪算法.通过建立一个参数可以灵活改善的多元统计模型来准确地估计各种相关信息,并通过最小化残差来自适应调整模型参数;在适度的噪声方差下根据拼贴距离找出最好的子树域中的近优父子树,通过使用四叉树分割来实现对噪声图像自适应地预测分形小波无噪图像编码,从而达到去除噪声和保护图像的边缘与细节的目的.实验结果表明,该方法在去除噪声的同时,能有效地保持图像的边缘特征和保留图像的精细结构.由于采用了预测小波分形编码,优化了算法结构,算法的处理速度比较快.因此,完全可以达到实时图像处理过程中的去噪预处理对处理速度的要求.In order to overcome the defects that fractal wavelet denoising method can＇t effectively protect the edges and details of image,a adaptive fractal wavelet image denoising scheme based on multivariate statistical model is presented in this article.A multivariate statistical model whose parameters can be improved flexibly is established to accurately estimate all relevant information in this method and the parameters of model are adaptively adjusted by minimizing the residual.In order to remove noise and protect the edges and details of image,near optimal parent subtree is found in the best subtree domain by using collage distance under moderate noise and quadtree segmentation is used to adaptive predict fractal wavelet coding of noiseless image from the noise image.Experimental results show that this method can effectively maintain the edge features of image and retain the fine structure of image primely while noise be removed.As the prediction of wavelet fractal coding is adopted in this scheme,the algorithm structure has been optimized,and the processing speed for algorithm is faster.Therefore,it can achieve the processing speed requirements for real time image denoising.

关 键 词：图像去噪 多元统计模型 分形小波变换 拼贴编码 RMSE和PSNR 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]