pharris

Performs Harris corner detection.

Synopsis

pharris sigma kappa [-m mask] [im_in|-] [im_out|-]

Description

pharris is a corner detector. Corners are T, L or Y junctions or points with strong texture variation. Corners correspond to double discontinuities of the intensity function caused by discontinuities in the reflectance or the depth functions.

The principle of the algorithm is to compute the covariance matrix C(x,y):

     | S(Ix.Ix) S(Ix.Iy) |
   C=|                   |
     | S(Ix.Iy) S(Iy.Iy) |

where Ix and Iy are image gradient components and S(x) is the sum symbol. The eigenvalues represent the major and minor axis of the elliptical approximation of the gradient vector distribution. If the smaller eigenvalue of the matrix is positive, it is considered as a corner.

To avoid to compute the eigenvalue, Harris proposes to compute the response function R(x,y) for each pixel:

   R=Ixx*Iyy-Ixy*Ixy - kappa * (Ixx+Iyy) * (Ixx+Iyy)

and than to choose the interest point as local maxima of function R(x,y).

kappa is a tunable parameter which determines how 'edge-phobic' the algorithm is. The value has to be determined empirically. In the literature values in the range 0.04 - 0.15 have been reported as feasible.

The size of research area for the local maxima is determined from the sigma parameter (size=6*sigma).

The result image is a float image that encodes for each pixel the strength of the response.

Parameters

sigma is the standard variation of the gaussian. It is used to define the size of neighborhood where local maxima are extracted (size=6*sigma). A typical value is between [1..3].
kappa determines how 'edge-phobic' the algorithm is. The typical value estimated by Harris is 0.04.

Inputs

im_in: a 2D image.

Outputs

im_dest: a 2D Float image.

Result

Returns SUCCESS or FAILURE.

Examples

Extracts corners from image tangram.pan and superimposes corners on the initial image.

   pharris 2 0.04 tangram.pan a.pan
   pbinarization 1e4 1e30 a.pan b.pan
   padd b.pan tangram.pan out.pan

C++ prototype

Errc PHarris( const Img2duc &im_in, Img2dsf &im_out, float sigma, float kappa );

Version française

Détection de points d'intérêt selon l'algorithme de Harris-Stephens.

Author: Régis Clouard