We present a demo on the morphological corner detector
developed by L.Alvarez, C.Cuenca and L.Mazorra
in the paper "Morphological Corner Detection. Application to Camera Calibration". (PS file) (PDF file) .
In this paper we study the application of the Affine Morphological Scale Space (AMSS) to corner detection with subpixel precision. Corner detection techniques are in general very sensitive to noise, so some kind of filtering is usually needed to remove noise before estimating the corner location, however, the filtering procedure changes the location of the corner, so the filtering introduce errors in the corner location. If we use the AMSS scale space as filtering, we can solve this problem because following the evolution of the extrema of the curvature across the scales we can recover the original location of the corners with subpixel precision. For more information about the AMSS scale space see the demo page Morphological Multiscale Analysis DEMO page .
Figure 1. Illustration of the Morphological Corner detector.
On the left, the
original real image which corresponds to a rectangle of a calibration object.
In the center, the result at the scale t=6, of the evolution of the original image
under the action of the affine invariant morphological multiscale analysis. On
the right, the evolution of the local extrema of the curvature across the scales
(In white, the original location of the extrema of curvature, in red and green,
the location of the extrema of curvatures across the scales). If you click HERE
you will see a movie a movie where you can see the evolution of the curvature
extrema across the scales.
Robust morphological corner detector algorithm.
(1) We compute, using AMSS, the image at the scale tn=t0+ndt where dt represents the discretization step for the scale and t0 represents the initial scale that we use to begin to look for corners.
(2) We compute for the scale t0 the location of the extrema of the curvature that we denote by (x0i,y0i), these points represent the initial candidates to be corners. We follow across the scales the location (xni,yni) of the curvature extrema.
(3) For each sequence (xni,yni), we compute in a robust way (using orthogonal regression and eliminating outliers) the best line which fit the sequence of points, this line corresponds to the bisector line of the corner. We compute also in a robust way the best line passing for the points (tn-t0,dist((xni,yni),(x0i,y0i))) The slope of such line, that we denote by (mi, determines the angle of the corner. We also estimate, the direction of the bisector line that we denote by (bxi,byi).
(4) Finally, we estimate the location (xi,yi) of the corner in the original image as:
Figure 2. We illustrate the influence of the initial scale t0 in order to remove the
spurious local extrema of curvature produces by the noise: On the left the evolution
of extrema of curvature for t0=0. On the right the evolution of extrema of curvature for t0=1.
Another advantage of the morphological corner detector is that you do not need to have a perfect corner given by the interception of 2 lines because you look for curvature extrema, in the next figure we show a corner detection illustration on a footbal.
Figure 3. Illustration of the Morphological Corner detector on
In blue, we have the recovered corners.
You can test freely the morphological corner detector by your own with just 2 limitations:
1. You can not use it for commercial purposes.
2. You have to acknowledge to the authors in any publications where you use the algorithm.
In order to get the C-source code you have to download the following files:
1. ami.h (some macros
that we use)
2. ami_calibracion (the library where the main function are defined )
3. ami_morphological_corner_detector_test.c (an example of application of the algorithm)
4. XMW_ami_morphological_corner_detector_test.c (A graphic windows implementation of the algorithm in the XMegaWave Software. You need to install XMegaWave to use this function see XMegaWave Installation for more details.
5. XMW_ami_local_morphological_corner_detector_test.c (A graphich window implementation of the morphological corner detector where we improve the location of a number of initial corners provided by the user by clicking on the image (tipically this tool is used for image calibration purposes). You need to install XMegaWave to use this function see XMegaWave Installation for more details.
when you execute
you will obtain in a file the information about the corner location and some corner quality parameters.