Has anyone implemented moravec corner detector in matlab??

This is what i have tried.But with very less accurate.

    clear all

close all

clc

tic

I=imread('panorama_image1.jpg');

originalmap=I;

w=5;

m=floor(size(I,1)/w);

n=floor(size(I,2)/w);

c=ceil(w/2)+(0:n-1)*w;

r=ceil(w/2)+(0:m-1)*w;

step=floor(w/2);

for y_unit=1:m

    for x_unit=1:n 

        v1=0;v2=0;v3=0;v4=0; 

        for i=-step:step-1 

            v1=v1+(I(r(y_unit),c(x_unit)+i)-I(r(y_unit),c(x_unit)+i+1))^2; 

            v2=v2+(I(r(y_unit)+i,c(x_unit)+i)-I(r(y_unit)+i+1,c(x_unit)+i+1))^2; 

            v3=v3+(I(r(y_unit)+i,c(x_unit))-I(r(y_unit)+i+1,c(x_unit)))^2; 

            v4=v4+(I(r(y_unit)-i,c(x_unit)+i)-I(r(y_unit)-i-1,c(x_unit)+i+1))^2; 
        end 

        IV_cr(y_unit,x_unit)=min([v1,v2,v3,v4]); 
    end 
end

exper_thr=20;

IV_cr(IV_cr<exper_thr)=NaN;

wf=9;

mf=floor(m/wf);

nf=floor(n/wf);

xc=[];

yc=[]; for y_unit=1:mf

    for x_unit=1:nf 

        [C,I]=max(IV_cr((y_unit-1)*wf+1:y_unit*wf,(x_unit-1)*wf+1:x_unit*wf))

        [C1,I1]=max(C); 

        IV_crch(y_unit,x_unit)=C1;
        row=I(I1); 

        col=I1; 

        crch_row(y_unit,x_unit)=(y_unit-1)*wf+row;

        crch_col(y_unit,x_unit)=(x_unit-1)*wf+col;

        yc=[yc,r((y_unit-1)*wf+row)];xc=[xc,c((x_unit-1)*wf+col)]; 

        IV_cr((y_unit-1)*wf+1:y_unit*wf,(x_unit-1)*wf+1:x_unit*wf)=NaN; 

        IV_cr((y_unit-1)*wf+row,(x_unit-1)*wf+col)=C1; 
    end 
end 
figure(1) 
imshow(originalmap)

figure(2)

plot(xc,yc,'*')

view(0,-90)

toc;

t=toc;

disp(['amount of time',num2str(t),'sec'])

figure(3)

imshow(originalmap)

hold on

plot(xc,yc,'R*')

axis on

end

Answer 1

[Thorsten]

I just did if for you:

function C = moravec(I)
%MORAVEC Moravec corner detector
%
%   C = MORAVEC(I)
%
%   [e-mail]  2015-05-20
if nargin == 0 % simple test
  I = zeros(100);
  I(40:60, 40:60) = 1;
  C = moravec(I);
  imshow(0.5*C+I, [])
  if nargout == 0, clear C, end
  return
end
D(:,:,1) = diff([I I(:,end)]')';
D(:,:,2) = diff([I(:,1) I]')';
D(:,:,3) = diff([I; I(end,:)]);
D(:,:,4) = diff([I(1,:); I]);
C = sum(D.^2, 3);
if nargout == 0, clear C, end

[B.k Sumedha]

Dimensions of matrices being concatenated are not consistent.??

[B.k Sumedha]

At line number 16

[Thorsten]

So far it works just for grayscale images, not color images, but you could extend it.

[B.k Sumedha]

Tried for grayscale image also.

[B.k Sumedha]

But did not help.

[Thomas Koelen]

You might have an alpha channel with your grayscale image, check:

size(image) and check if it says (x x 2) or (x x 1), if it says (x x 2) you have to remove the alpha channel first.

[B.k Sumedha]

It says 225 300

[B.k Sumedha]

How to remove that??

[Thorsten]

I just tried the following, which works fine:

>> I = imread('cameraman.tif');
>> C = moravec(I);
>> size(I)
ans =
   256   256

Can you reproduce that?

I also tried

 R = rand(225, 300); C = moravec(R);

which also works fine for me.

[B.k Sumedha]

Bad luck!!! Not happening here. Can u display the output here.

[Thorsten]

What do you mean by "Bad luck"? Does the result look strange, or do you get a Matlab error? The Moravec detector is highly sensitive to noise, so you get fine results for artificial images (run moravec without arguments to see an overlay of the image and the detected corners), but less so for natural images.

[B.k Sumedha]

ok fyn

Answer 2

[Ignacio Rocco]

Hi,

Thanks Thorsten for your code! However I think it's missing a few details. I slightly modified the implementation according to the original text by Moravec [1980], and I think now it's producing much better results.

Here's the updated code.

function C = moravec(I)
%MORAVEC Moravec corner detector
%
%   C = MORAVEC(I)
%
%   [e-mail]  2015-05-20
%   [e-mail]                  2019-11-14

if nargin == 0 % simple test
  I = zeros(100);
  I(40:60, 40:60) = 1;
  C = moravec(I);
  imshow(0.5*C+I, [])
  if nargout == 0, clear C, end
  return
end
D(:,:,1) = conv2(conv2(single(I),[1 -1],'same').^2,ones(3),'same');
D(:,:,2) = conv2(conv2(single(I),[1;-1],'same').^2,ones(3),'same');
D(:,:,3) = conv2(conv2(single(I),[1 0;0 -1],'same').^2,ones(3),'same');
D(:,:,4) = conv2(conv2(single(I),[0 1;-1 0],'same').^2,ones(3),'same');
C = min(D.^2,[], 3);
if nargout == 0, clear C, end