非局部均值滤波

通过使用非局部均值滤波可以实现对自然图像的去噪，效果较好

function [output]=NLmeansfilter（inputtfh）
 
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 %
 %  input: image to be filtered
 %  t: radio of search window
 %  f: radio of similarity window
 %  h: degree of filtering
 %
 %  Author: Jose Vicente Manjon Herrera & Antoni Buades
 %  Date: 09-03-2006
 %
 %  Implementation of the Non local filter proposed for A. Buades B. Coll and J.M. Morel in
 %  “A non-local algorithm for image denoising“
 %
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

 % Size of the image
 [m n]=size（input）;
 
 
 % Memory for the output
 Output=zeros（mn）;

 % Replicate the boundaries of the input image
 input2 = padarray（input[f f]‘symmetric‘）;
 
 % Used kernel
 kernel = make_kernel（f）;
 
 kernel = kernel / sum（sum（kernel））;
 
 
 for i=1:m
     for j=1:n
                 
         i1 = i+ f;
         j1 = j+ f;
        
         
         W1= input2（i1-f:i1+f  j1-f:j1+f）;
         
         wmax=0; 
         average=0;
         sweight=0;
         
         rmin = max（i1-tf+1）;
         rmax = min（i1+tm+f）;
         smin = max（j1-tf+1）;
         smax = min（j1+tn+f）;
         
         for r=rmin:1:rmax
             for s=smin:1:smax
                              
                 
                 if（r==i1 && s==j1） continue; end;
                
                 
                 W2= input2（r-f:r+f  s-f:s+f）;                
                 
                 d = sum（sum（kernel.*（W1-W2）.*（W1-W2）））;
                              
                 
                 w=exp（-d/h）;
                 
                                 
                 if w>wmax
                
                    wmax=w;
                   
                end
                
                sweight = sweight + w;
                average = average + w*input2（rs）;
                                  
             end
 
         end
             
        average = average + wmax*input2（i1j1）;
        sweight = sweight + wmax;
        
           
        if sweight > 0
            output（ij） = average / sweight;
        else
            output（ij） = input（ij）;
        end
        
        
     end
 end
 
 
    
 
 function [kernel] = make_kernel（f）
        
        
 
    kernel=zeros（2*f+12*f+1）;

    for d=1:f
    
        value= 1 / （2*d+1）^2 ;
    
        for i=-d:d
            for j=-d:d
                kernel（f+1-if+1-j）= kernel（f+1-if+1-j） + value ;
        end
    end
end


kernel = kernel ./ f

        

 属性            大小     日期    时间   名称
----------- ---------  ---------- -----  ----
     文件        2623  2007-09-15 00:51  NLmeansfilter.m