细化算法matlab

最简单的方法

最简单的方法是用matlab自带的函数：
BW2 = bwmorph(BW,'thin',n);
n是要细化迭代的次数，也可以是Inf(没有引号)。Inf表示算法会一直迭代直到图像不再改变。
算法原理参考 [1]。

稍微复杂点的方法

Zhang-Suen算法
原理简介：http://www.cnblogs.com/mikewolf2002/p/3321732.html
算法实现：http://www.chinabaike.com/t/9642/2014/0624/2547871.html
将下列三个m文件和主m文件放到一起，主m文件这样调用:BW2=zs(BW1);
zs.m

%zs.m function out=zs(im) % %zs appises the Zhang-Suen skeletonization algorithm to image IM. IM must %be binary. % luteven=makelut('zseven',3); lutodd=makelut('zsodd',3); done=0; N=2; last=im; previous=applylut(last ,lutodd); current=applylut(previous,luteven); while done==0, if all(current(:)==last(:)), done=1; end N=N+1; last=previous; previous=current; if mod(N,2)==0, current=applylut(current,luteven); else current=applylut(current,lutodd); end; end; out=current; end

zseven.m

%zseven.m function out=zseven(nbhd) s=sum(nbhd(:))-nbhd(5); temp1=(2<=s)&&(s<=6); p=[nbhd(1) nbhd(4) nbhd(7) nbhd(8) nbhd(9) nbhd(6) nbhd(3) nbhd(2)]; pp=[p(2:8) p(1)]; xp=sum((1-p).*pp); temp2=(xp==1); prod1=nbhd(4)*nbhd(8)*nbhd(2); prod2=nbhd(4)*nbhd(6)*nbhd(2); temp3=(prod1==0)&&(prod2==0); if temp1&temp2&temp3&nbhd(5)==1 out=0; else out=nbhd(5); end; end

zsodd.m

%zsodd.m function out=zsodd(nbhd) s=sum(nbhd(:))-nbhd(5); temp1=(2<=s)&&(s<=6); p=[nbhd(1) nbhd(4) nbhd(7) nbhd(8) nbhd(9) nbhd(6) nbhd(3) nbhd(2)]; pp=[p(2:8) p(1)]; xp=sum((1-p).*pp); temp2=(xp==1); prod1=nbhd(4)*nbhd(8)*nbhd(6); prod2=nbhd(8)*nbhd(6)*nbhd(2); temp3=(prod1==0)&&(prod2==0); if temp1&temp2&temp3&nbhd(5)==1 out=0; else out=nbhd(5); end; end

自己折腾的过程

参考[2]中骨架化的算法，不知道是算法写错了，还是本来论文就没有论述清楚，总是会把左上角给弄断，先记在下面，以后再改吧。

Isrc = imread('pic/0.bmp'); if ndims(Isrc) == 3 I = rgb2gray(Isrc); else I = Isrc; end I = im2bw(I,0.5); %外围扩展边界 [H,W] = size(I); I = [zeros(2,W);I;zeros(2,W)]; I = [zeros(H+4,2),I,zeros(H+4,2)]; [H,W] = size(I); for k = 1:10 %迭代次数 p = zeros(8,1); NzI = zeros(H,W); ZI = zeros(H,W); P024I = zeros(H,W); P246I = zeros(H,W); I2 = I; for i = 2:H-1 for j = 2:W-1 if I(i,j) p(1) = I(i,j+1); p(2) = I(i-1,j+1); p(3) = I(i-1,j); p(4) = I(i-1,j-1); p(5) = I(i,j-1); p(6) = I(i+1,j-1); p(7) = I(i+1,j); p(8) = I(i+1,j+1); NzI(i,j) = sum(p); ZI(i,j) = sum(abs(diff(p)))+abs(p(8)-p(1)); P024I(i,j) = p(1) * p(3) * p(5); P246I(i,j) = p(3) * p(5) * p(7); end end end for i = 3:H-1 for j = 3:W-1 if I(i,j) if ((NzI(i,j) >= 2) && (NzI(i,j) <= 6)) && ... (ZI(i,j) == 2) && ... ((P024I(i,j)==0)||(ZI(i,j-1)~=2)) && ... ((P246I(i,j)==0)||(ZI(i-1,j)~=2)) I2(i,j) = 0; end end end end I = I2; end figure(1); subplot(121); imshow(Isrc); subplot(122); imshow(I2);

[1]: Lam, L., Seong-Whan Lee, and Ching Y. Suen, “Thinning Methodologies-A Comprehensive Survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol 14, No. 9, September 1992, page 879, bottom of first column through top of second column.
[2]周凤香. 工业生产线标签字符识别系统的设计与实现[D].电子科技大学,2013.