Low-level feature extraction (including edge detection) - Feature Extraction and Image Processing - page 144

first part of the code computes the coefficients of the matrix M according to Equation 4.56.
Then, these values are used in the curvature computation.

%Harris Corner Detector
%op=H Harris
%op=M Minimum direction
function outputimage=Harris(inputimage, op)

w=4; %Window size=2w+1

k=100; %Second term constant

[rows, columns]=size(inputimage); %Image size

outputimage=zeros(rows, columns); %Result image

[difx, dify]=Gradient(inputimage); %Differential

[M,A]=Edges(inputimage); %Edge Suppression

M=MaxSupr(M,A);

%compute correlation
for x=w+1: columns-w %pixel (x,y)

for y=w+1:rows-w
if M(y, x)~=0
%compute window average
A=0;B=0;C=0;
for i=-w:w
for j=-w:w
A=A+difx(y+i,x+j)^2;
B=B+dify(y+i,x+j)^2;
C=C+difx(y+i,x+j)*dify(y+i,x+j);
end
end

if(op== ′ H ′ )
outputimage(y,x)=A*B-C^2-k*(A+B);

else

dx=difx(y,x);
dy=dify(y,x);

if dx*dx+dy*dy~=0
outputimage(y,x)=((A*dy*Dy-
2*C*dx*dy+B*dx*dx)/(dx*dx+dy*dy));
end
end
end
end
end

Code 4.19 Curvature by autocorrelation

Figure 4.35 shows the results of computing curvature using this implementation. The
results are capable of showing the different curvature along the border. We can observe that

k ( x , y ) produces more contrast between lines with low and high curvature than κ u , v ( x , y ).