# 角点检测

E(u,v)的椭圆形式如下图：

Harris角点检测算法就是对角点响应函数R进行阈值处理：R > threshold，即提取R的局部极大值。

Shi--Tomasi角点检测法，如果像素点的最小特征值大于周围像素的特征值，则该点是角点。

<span style="font-size:18px;"><strong>im=imread('lena.jpg');
tau=100;
im=double(im);
keyXs=[];
keyYs=[];
win=3;
[height,width] = size(im);
result = zeros(height,width);
%Then I will get the gradients of the image along the x and y axises.
sobel_x=1/4*[-1 0 1;-2 0 2;-1 0 1];
sobel_y=1/4*[-1 0 1;-2 0 2;-1 0 1]';
diffx=imfilter(im,sobel_x);         %对图像x方向进行梯度
diffy=imfilter(im,sobel_y);       %对图像y方向的梯度进行计算
%For smoothing the differentiation of the image along the x and y
%direction, the gauss filter of the diffx and diffy is must.
gauss_win=win;
sigma=1;
[x,y]=meshgrid(-gauss_win:gauss_win,-gauss_win:gauss_win);
gauss2D=exp(-(x.^2+y.^2)/(2*sigma.^2));  %产生高斯算子
gauss2D=gauss2D/(sum(sum(gauss2D)));  %对高斯算子进行归一化
%Then calculate the M matrix.
A=imfilter(diffx.*diffx,gauss2D);      %二阶x方向梯度进行高斯滤波
B=imfilter(diffy.*diffy,gauss2D);      %二阶y方向梯度进行高斯滤波
C=imfilter(diffx.*diffy,gauss2D);      %对图像x y方向的梯度进行高斯滤波
supress_win=2;
threshold=100;
points_count=0;
bigger=zeros(height,width);
smaller=zeros(height,width);
for x=1:width
for y=1:height
M=[A(y,x) C(y,x);C(y,x) B(y,x)];
%It is too time-consuming.
%eigenvalue=eig(M);
%bigger(y,x)=max(eigenvalue);
%smaller(y,x)=min(eigenvalue);
temp1=M(1,1)+M(2,2);
temp2=sqrt((M(1,1)-M(2,2))^2+4*M(1,2)^2);
bigger(y,x)=(temp1+temp2)/2;
smaller(y,x)=(temp1-temp2)/2;
end
end
for x=supress_win+1:width-supress_win
for y=supress_win+1:height-supress_win
temp=smaller(y,x);
if(temp>threshold)
%Then I will make the non-maximumu suppression to the
%samller matrix after the threholding.
flag=0;
for i=-supress_win:supress_win
for j=-supress_win:supress_win
if(temp>=smaller(y+j,x+i))
flag=flag+1;
end
end
end
if(flag==((2*supress_win+1)*(2*supress_win+1)))
result(y,x)=1;
points_count=points_count+1;
keyXs(points_count)=x;
keyYs(points_count)=y;
end
end
end
end
end
</strong></span>

©️2019 CSDN 皮肤主题: 大白 设计师: CSDN官方博客