canny边缘检测算法的matlab实现

function result=canny(I,sz,T1,T2) % perform canny edge detector. I is the input image, sz indicates the size % of gaussian, T1 is the lower threshold and T2 is the upper threshold. % smooth the image with a gaussian filter h=fspecial('gaussian',sz,sz/6); I=imfilter(I,h,'replicate'); I=double(I); % choose priwitt to compute the gradient magnitude image g(x,y) GX=[-1 -1 -1; 0 0 0; 1 1 1] ;% Gx GY=[-1 0 1;-1 0 1; -1 0 1]; % Gy I1=imfilter(I,GX,'replicate'); I2=imfilter(I,GY,'replicate'); g=sqrt(I1.^2+I2.^2); g [x,y]=size(I); % get the size of I. % compute the edge direction image theta(x,y) theta=zeros(x,y); for i=1:x for j=1:y if(I1(i,j)==0 && I2(i,j)==0) % if both Gx and Gy are equal to 0, set the direction to be 0 theta(i,j)=0; else theta(i,j)=atan(I2(i,j)./I1(i,j))*180/3.1415926; % get the direction in degrees. end % set the direction's range to be [0,360] if(I1(i,j)>=0&&I2(i,j)<0) theta(i,j)=theta(i,j)+360; elseif(I1(i,j)<0&&I2(i,j)>=0) theta(i,j)=theta(i,j)+180; elseif(I1(i,j)<0&&I2(i,j)<0) theta(i,j)=theta(i,j)+180; end end end % quantize the angle to the nearest angle:{0,45,90,135,180,225,270,315,360} angle=(0:8)*45; for i=1:x for j=1:y for k=1:9 if(abs(theta(i,j)-angle(k))<=45/2) theta(i,j)=angle(k); end end end end % call localmax function to perform the nonmaximal supression G=localmax(g,theta); G=uint8(G); G(find(G>T2*max(G(:))))=255; % change the pixels above upper threshold to be edges. G(find(G<T1*max(G(:))))=0; % change the pixels below lower threshold to be background. % deal with the weak edges. if a weak edge pixel is adjacent to a edge pixel, mark the pixel as an edge pixel. % repeat this for 20 times and then set the rest weak edge pixels to be background. n=0; while(size(find(G>0 & G<255))~=[1 0]) for i=2:x-1 for j=2:y-1 if (G(i,j)>0 && G(i,j)<255) if (G(i-1,j-1)==255 || G(i-1,j)==255 || G(i-1,j+1)==255 || (G(i,j-1)==255) || (G(i,j+1)==255) ||(G(i+1,j-1)==255) || G(i+1,j)==255 || G(i+1,j+1)==255) G(i,j)=255; end end end end n=n+1; if(n==20) break; end end G(find(G>0 & G<255))=0; result=im2bw(G); % change the image, which only has intensity of 0 and 255, to a binary image. end function result=localmax(M,Theta) % perform the nonmaximal supression. M is the input gradient magnitude % image, Theta is the edge direction image. [x,y]=size(M); result=M; for i=3:x-2 for j=3:y-2 m=0; % if the gradient magnitude is smaller than that of the adjacent % pixels along the quantized direction, set the magnitude to 0 switch Theta(i,j)/45 case {0,4,8} m=max([M(i,j),M(i,j+1),M(i,j-1)]); case {1,5} m=max([M(i,j),M(i-1,j+1),M(i+1,j-1)]); case {2,6} m=max([M(i,j),M(i-1,j),M(i+1,j)]); case {3,7} m=max([M(i,j),M(i-1,j-1),M(i+1,j+1)]); end if(result(i,j)<m) result(i,j)=0; end end end end