IM_HARRIS

Fixed mapping executing the Harris corner detector

     X = IM_HARRIS(A,N,SIGMA)
     X = A*IM_HARRIS([],N,SIGMA)
     X = A*IM_HARRIS(N,SIGMA)

Input
A	Datafile or dataset with images
N	Number of desired Harris points per image (default 100)
SIGMA	Smoothing size (default 3)

Output
X	Dataset with a [N,3] array with for every image  x, y and strength per Harris point.

Description

We use Kosevi's [1] software to find the corner points according to  Harris [2]. On top of the Kosevi Harris point detector we run

• multi-feature images (e.g. color images) are averaged
• only points that are maximum in a K x K window are selected.  If less points are found, K is iteratively reduced.  The initial value of K is about 4*SIGMA.  Although SIGMA can be interpreted as scaling parameter, it might be  better to appropriately subsample images instead of using a large SIGMA.

If you use this software for publications, please refer to [1] and [2].

Reference(s)

[1] P. D. Kovesi, MATLAB and Octave Functions for Computer Vision and Image Processing, School of Computer Science && Software Engineering,  The University of Western Australia. Available from: . [2] C. Harris and M. Stephens, A combined corner and edge detector, Proc. 4th Alvey Vision Conf., 1988, pp. 147-151.

Example(s)

 delfigs
 a = kimia;                  % take simple shapesas example
 b = gendat(a,25)*im_gray;   % just 25 images at random
 c = data2im(b);             % convert dataset to images for display
 x = im_harris(b,15,1);      % compute maximum 15 Harris points at scale 1
 y = data2im(x);             % unpack dataset with results 
 for j=1:25                  % show results one by one
     figure(j); imagesc(c(:,:,1,j)); colormap gray; hold on
     scatter(y(:,1,1,j),y(:,2,1,j),'r*');
 end
 showfigs