% 原始非局部均值(Original Non-local Means, ONLM)算法，基本思想是利用图像中大量冗余信息，对图像中所有相似邻域内像素的灰度值

% 根据权重系数计算加权平均得到受污染像素的灰度估计值，其中权重系数由加权核函数和像素邻域之间的加权欧式距离决定。由于NLM 算法

% 计算量非常大，为提高算法的效率，采用限定邻域搜索范围的 NLM 算法,像素邻域大小 7×7，邻域搜索范围 21×21，并且调整核函数的滤

% 波参数使去噪性能达到最佳

clear;

% close all;

clc;

searchBox_r=10;%搜索框半径

similarBox_r=3;%相似框半径

h=20;%指数函数的衰减因子

rawImage=imread('barbara512.bmp');

% rawImage=imresize(rawImage,0.2);

[rawImage_m, rawImage_n]=size(rawImage);

denoisedImage=zeros(rawImage_m,rawImage_n);

% figure

% imshow(rawImage,[]);%显示原始图像

% title('raw image')

% noise=wgn(rawImage_m,rawImage_n,25);

% noiseImage=double(rawImage)+double(noise);

noiseImage=imnoise(rawImage,'gaussian', 0, 10^2/255^2);

H1=figure;

imshow(noiseImage,[]);%显示加入噪声后图像

title('noise image');

padNoiseImage = padarray(noiseImage,[searchBox_r searchBox_r],'symmetric'); %将边缘对称折叠上去

padNoiseImage = double(padNoiseImage);

%计算高斯核

alpha=1;%高斯加权欧氏距离的标准差

gaussMat = repmat(-similarBox_r:similarBox_r,2*similarBox_r+1,1);

gaussWeight = exp(-((gaussMat).^2+(gaussMat').^2)/(2*alpha*alpha));

%像素点遍历(像素点在搜索区域中心)

for pixel_i=searchBox_r+1:rawImage_m+searchBox_r

for pixel_j=searchBox_r+1:rawImage_n+searchBox_r

average=0;

sweight=0;

NI= padNoiseImage(pixel_i-similarBox_r:pixel_i+similarBox_r,pixel_j-similarBox_r:pixel_j+similarBox_r);%取出像素点邻域

%相似框在搜索框中遍历

for similarBox_i=pixel_i-searchBox_r+similarBox_r:pixel_i+searchBox_r-similarBox_r

for similarBox_j=pixel_j-searchBox_r+similarBox_r:pixel_j+searchBox_r-similarBox_r

NJ= padNoiseImage(similarBox_i-similarBox_r:similarBox_i+similarBox_r,similarBox_j-similarBox_r:similarBox_j+similarBox_r);%取出相似框中心点邻域

difMatrix=(NI-NJ).*gaussWeight;%漏乘了高斯加权矩阵

dij=sum(sum(difMatrix.^2));%代表两个矩阵的相关性,利用像素点的相关性来还原图像

fk=exp(-dij/(h^2));

sweight=sweight+fk;%计算权重和

average = average + fk*padNoiseImage(similarBox_i,similarBox_j); %计算加权和

end

end

denoisedImage(pixel_i-searchBox_r,pixel_j-searchBox_r) = average / sweight;%求加权均值

end

end

H2=figure;

imshow(denoisedImage,[]);%滤波后图像

title('NLM denoised image');

figure(H1);

text(0,0,sprintf('PSNR:%f',my_psnr(double(rawImage),double(noiseImage))));

figure(H2);

text(0,0,sprintf('PSNR:%f',my_psnr(double(rawImage),double(denoisedImage))));

一键复制

编辑

Web IDE

原始数据

按行查看

历史