快速均值滤波

采用积分图进行加速，实现与滤波窗口大小无关的效率

积分图:图像I的积分图S是与其大小相同的图像，S的每一像素S(u,v)存贮的是I(u,v)左上角所有像素的颜色值之和。

积分图可增量计算，只需对原图进行一遍扫描:S(u,v)=S(u,v-1)+sum(I(1:u,v))

设滤波窗口大小为 2w+1 ，滤波结果为图像 O ，则：

Z=(2w+1)*(2w+1)为像素个数；中括号内即为滤波窗口覆盖的像素颜色值之和；

clear;
A1=imread('H:/数字图像处理/13_input.jpg');
w=str2double(inputdlg('请输入滤波窗口大小', 'INPUT scale factor', 1, {'2'}));
z=(2*w+1)*(2*w+1);
A2=imnoise(A1,'salt & pepper',0.01);
%A2=rgb2gray(A2);
[nrows,ncols,ncoms]=size(A2);
OUT = uint8(zeros(nrows,ncols,ncoms));%输出图像
if(ncoms==1)
    S=zeros(nrows,ncols);
    S(1,1)=A2(1,1);
    for u=2:nrows
        S(u,1)=S(1,1)+A2(u,1);
    end    
    for u=1:nrows
       for v=2:ncols
         S(u,v)=S(u,v-1)+sum(A2(1:u,v));          
       end
    end
    t=S(3,2)/6;
  for v=1:ncols-w
      for i=1:w+1
        for j=0:w
            OUT(i,v+j)=S(i,v+j)/(i*(v+j));            
        end        
      end    
  end
for v=1:ncols-w+1
    for i=nrows-w+1:nrows
        for j=0:w-1
           OUT(i,v+j)=S(i,v+j)/(i*(v+j));            
        end        
    end    
end
for u=w+2:nrows-w-w
    for i=0:w
        for j=1:w+1
            OUT(u+i,j)=S(u+i,j)/((u+i)*j);            
        end
    end
end
for u=w+2:nrows-w-w+1
    for i=0:w-1
        for j=ncols-w+1:ncols
            OUT(u+i,j)=S(u+i,j)/((u+i)*j);            
        end
    end
end
for u=w+2:nrows-w
    for v=w+2:ncols-w
        OUT(u,v)=(1/z)*(S(u+w,v+w)+S(u-w-1,v-w-1)-S(u+w,v-w-1)-S(u-w-1,v+w));        
    end
end
end


if(ncoms==3)
R = A2(:,:,1);
G = A2(:,:,2);
B = A2(:,:,3);
%输出图像的RGB分量
RR = uint8(zeros(nrows,ncols));
GG = uint8(zeros(nrows,ncols));
BB = uint8(zeros(nrows,ncols));
%各分量上的积分量
Sr =zeros(nrows,ncols);
Sg =zeros(nrows,ncols);
Sb =zeros(nrows,ncols);
Sr(1,1)=R(1,1);
Sg(1,1)=G(1,1);
Sb(1,1)=B(1,1);
for u=2:nrows
    Sr(u,1)=Sr(u-1,1)+R(u,1);
    Sg(u,1)=Sg(u-1,1)+G(u,1);
    Sb(u,1)=Sb(u-1,1)+B(u,1);
end
for u=1:nrows
    for v=2:ncols
        Sr(u,v)=Sr(u,v-1)+sum(R(1:u,v));
        Sg(u,v)=Sg(u,v-1)+sum(G(1:u,v));
        Sb(u,v)=Sb(u,v-1)+sum(B(1:u,v));        
    end
end
%边界处理
for v=1:ncols-w
    for i=1:w+1
        for j=0:w
            RR(i,v+j)=Sr(i,v+j)/(i*(v+j));
            GG(i,v+j)=Sg(i,v+j)/(i*(v+j));
            BB(i,v+j)=Sb(i,v+j)/(i*(v+j));            
        end        
    end    
end
for v=1:ncols-w+1
    for i=nrows-w+1:nrows
        for j=0:w-1
            RR(i,v+j)=Sr(i,v+j)/(i*(v+j));
            GG(i,v+j)=Sg(i,v+j)/(i*(v+j));
            BB(i,v+j)=Sb(i,v+j)/(i*(v+j));
        end        
    end    
end
for u=w+2:nrows-w-w
    for i=0:w
        for j=1:w+1
            RR(u+i,j)=Sr(u+i,j)/((u+i)*j);
            GG(u+i,j)=Sg(u+i,j)/((u+i)*j);
            BB(u+i,j)=Sb(u+i,j)/((u+i)*j);
        end
    end
end
for u=w+2:nrows-w-w+1
    for i=0:w-1
        for j=ncols-w+1:ncols
            RR(u+i,j)=Sr(u+i,j)/((u+i)*j);
            GG(u+i,j)=Sg(u+i,j)/((u+i)*j);
            BB(u+i,j)=Sb(u+i,j)/((u+i)*j);
        end
    end
end
for u=w+2:nrows-w
    for v=w+2:ncols-w
        RR(u,v)=(1/z)*(Sr(u+w,v+w)+Sr(u-w-1,v-w-1)-Sr(u+w,v-w-1)-Sr(u-w-1,v+w));
        GG(u,v)=(1/z)*(Sg(u+w,v+w)+Sg(u-w-1,v-w-1)-Sg(u+w,v-w-1)-Sg(u-w-1,v+w));
        BB(u,v)=(1/z)*(Sb(u+w,v+w)+Sb(u-w-1,v-w-1)-Sb(u+w,v-w-1)-Sb(u-w-1,v+w));
    end
end
OUT(:,:,1) = RR;
OUT(:,:,2) = GG;
OUT(:,:,3) = BB;
end
imshow(A1),title('原图');
figure;
imshow(A2),title('加噪声后的图片');
figure;
imshow(OUT),title('均值滤波后的图片');