HELP CORNERS_HS                                     David Young
                                                    October 2003

LIB * CORNERS_HS implements the Harris-Stephens (also known as the
"Harris" or "Plessey") corner detector. Harris, C. and Stephens, M.
(1988) A combined corner and edge detector, ACV88 Proceedings of the
Fourth Alvey Vision Conference, pp 147-151.

corner_response_hs(I, sigma, k, R) -> R
corner_response_hs(I, region, sigma, k, R) -> R
        Computes the corner response R for the image and parameters
        given. This is defined by the following equations:

                              2
                R = Det - k Tr

        where
                            2
                Det = AB - C
                Tr = A + B

                     2
                A = X  * G(sigma)
                     2
                B = Y  * G(sigma)
                C = (XY) * G(sigma)

                X = I * Dx
                Y = I * Dy

        and where "*" means the convolution operation, Dx means the
        [-1 0 1] mask and Dy its vertical equivalent, and G(sigma) means
        the Gaussian smoothing mask with width parameter sigma.

        The arguments are:

        I: the input image, an array containing numbers

        region: if present, a 4-element list specifying the region of
        the image from which to take data, in *boundslist style. If
        absent, the whole of the array is used.

        sigma: a positive number, the smoothing parameter

        k: a positive number, which controls sensitivity to edges or
        corners. The best value for an application should be found
        empirically - a reasonable starting point appears to be 0.04 for
        images with grey levels in the range 0 to 1, but Harris and
        Stephens fail to discuss this.

        R: on entry to the procedure this may be:

            * an array, which will be filled with the results and
            returned. Its boundslist must be the same as region, if
            given; otherwise its boundslist must be the same as that of
            I;

            * <false> - a new array will be created and returned;

            * anything else - an array will be created using
            oldsfloatarray (see *newsfloatarray) using the input
            argument as the tag. This will be returned.


corners_hs(I, sigma, k, thr) -> list
        Returns a list of corner positions. I, sigma and k are as for
        corner_response_hs above. thr is a number giving a threshold in
        R for local maximum detection. list is a list of possible corner
        positions, taken to be those local maxima of the response array
        that exceed the threshold, in the format used by *array_peaks.


--- $popvision/help/corners_hs
--- Copyright University of Sussex 2003. All rights reserved.