1 简介
提出了一种基于各向异性扩散的图像分割算法.对现有的各向异性扩散的正则化方法进行了分析.根据微分几何中共形映射的有关理论,把原扩散方程分解为关于表面曲率的二阶方程,给出了分解式的正则化条件,保证了解的稳定性.通过对扩散系数的调节,提高了对各向异性扩散过程的控制能力.在形态学分割的基础上,通过能量函数最小化实现非线性尺度空间中的区域合并,消除了分水岭算法造成的严重过分割现象.实验结果表明,该算法的分割结果可为后续识别和理解提供较理想的方式.
2 部分代码
function [Normalized,X,Y] = AffineIm1(im, a, b, c, d, point1, point2)%分别对wndim的四个角进行仿射变换%f1,f2,f3,f4分别是目标图像的四个顶点Mysize = size(im);f1 = [a,0;0,a]*([cos(pi/b),-sin(pi/b);sin(pi/b),cos(pi/b)]*[1-point1,1-point2]');f2 = [a,0;0,a]*([cos(pi/b),-sin(pi/b);sin(pi/b),cos(pi/b)]*[1-point1,Mysize(1,2)-point2]');f3 = [a,0;0,a]*([cos(pi/b),-sin(pi/b);sin(pi/b),cos(pi/b)]*[Mysize(1,1)-point1,1-point2]');f4 = [a,0;0,a]*([cos(pi/b),-sin(pi/b);sin(pi/b),cos(pi/b)]*[Mysize(1,1)-point1,Mysize(1,2)-point2]');f1(1,1) = round(f1(1,1)+c);f1(2,1) = round(f1(2,1)+d);f2(1,1) = round(f2(1,1)+c);f2(2,1) = round(f2(2,1)+d);f3(1,1) = round(f3(1,1)+c);f3(2,1) = round(f3(2,1)+d);f4(1,1) = round(f4(1,1)+c);f4(2,1) = round(f4(2,1)+d);%找出被仿射后图像的外接矩形minx = min([f1(1,1),f2(1,1),f3(1,1),f4(1,1)]);maxx = max([f1(1,1),f2(1,1),f3(1,1),f4(1,1)]);miny = min([f1(2,1),f2(2,1),f3(2,1),f4(2,1)]);maxy = max([f1(2,1),f2(2,1),f3(2,1),f4(2,1)]);flagx = 1;flagy = 1;for i = minx:maxx flagy = 1; for j = miny:maxy orco = inv([cos(pi/b),-sin(pi/b);sin(pi/b),cos(pi/b)])*([1/a,0;0,1/a]*([i,j]'-[c,d]'))+[point1,point2]'; if orco(1,1)<1 | orco(1,1)>Mysize(1,1) | orco(2,1)<1 | orco(2,1)>Mysize(1,2) Normalized(flagx,flagy) = 0; else m = floor(orco(1,1)); n = floor(orco(2,1)); u = orco(1,1)-m; v = orco(2,1)-n;% if u>0.5 & v>0.5% Normalized(flagx,flagy) = wndim(m+1,n+1);% elseif u>0.5 & v<0.5% Normalized(flagx,flagy) = wndim(m+1,n);% elseif u<0.5 & v>0.5% Normalized(flagx,flagy) = wndim(m,n+1);% elseif u<0.5 & v<0.5% Normalized(flagx,flagy) = wndim(m,n);% end Normalized(flagx,flagy) = round((1-u)*(1-v)*im(m,n) + (1-u)*v*im(m,n+1) + u*(1-v)*im(m+1,n) + u*v*im(m+1,n+1)); end flagy = flagy+1; end flagx = flagx+1;endX = abs(minx);Y = abs(miny);% mysize = min([maxx-minx,maxy-miny]);% hlfsize = floor((mysize+1)/2)-5;% hlfsize = 30;% Normalized = Normalized(Xw(1,1)-hlfsize:Xw(1,1)+hlfsize,Xw(2,1)-hlfsize:Xw(2,1)+hlfsize);% % Temp = Temp(Xw(1,1)-hlfsize:Xw(1,1)+hlfsize,Xw(2,1)-hlfsize:Xw(2,1)+hlfsize);% [w,h] = size(Normalized);% Xw = [floor(w/2)+1,floor(w/2)+1]';
3 仿真结果
4 参考文献
[1]彭启民, 贾云得. 基于各向异性扩散的图像分割算法[J]. 北京理工大学学报, 2005, 25(4):4.
