NL-Means

function DenoisedImg=NLmeans(I,ds,Ds,h)

%此程序是对单个像素做非局部去噪，并非块，

%当目标像素扩展的块与搜索到的块近似时认为搜索到的像素与目标像素相近，权重大

%I:含噪声图像

%ds:邻域窗口半径

%Ds:搜索窗口半径

%h:高斯函数平滑参数

%DenoisedImg：去噪图像

I=double(I);

[m,n]=size(I);

DenoisedImg=zeros(m,n);

PaddedImg = padarray(I,[ds,ds],'symmetric','both');

kernel=ones(2*ds+1,2*ds+1);

kernel=kernel./((2*ds+1)*(2*ds+1));

h2=h*h;

for i=1:m    %整幅图像挨着取

    for j=1:n

        i1=i+ds;

        j1=j+ds;

        W1=PaddedImg(i1-ds:i1+ds,j1-ds:j1+ds);%邻域窗口1

        wmax=0;

        average=0;

        sweight=0;

        %%搜索窗口

        rmin = max(i1-Ds,ds+1);

        rmax = min(i1+Ds,m+ds);

        smin = max(j1-Ds,ds+1);

        smax = min(j1+Ds,n+ds);

        for r=rmin:rmax

            for s=smin:smax

                if(r==i1&&s==j1)

                continue;

                end

                W2=PaddedImg(r-ds:r+ds,s-ds:s+ds);%邻域窗口2

                Dist2=sum(sum(kernel.*(W1-W2).*(W1-W2)));%邻域间距离

                w=exp(-Dist2/h2);  %距离越大，w越小

                if(w>wmax)

                    wmax=w;

                end

                sweight=sweight+w;%权重归一

                average=average+w*PaddedImg(r,s);%

            end

        end

        average=average+wmax*PaddedImg(i1,j1);%自身取最大权值

        sweight=sweight+wmax;

        DenoisedImg(i,j)=average/sweight;

    end

end

 

 

这里边有作者的笔记：https://blog.csdn.net/wxz1120171495/article/details/81488863

代码与这个相同：https://blog.csdn.net/u010839382/article/details/48229579