![](bg4.png)
continue it found all inliers after an average of about
32000 iterations.
Figure 8 shows both the RPM and the RDM for
the whole set as well as for the case when 5 outliers
are remaining. Clearly it is hard to see which rows
and columns to remove in the upper row, while it
becomes more and more clear as the outliers are
removed.
Figure 8: The RPM matrix is visualized showing one
symmetric matrix for , to the left and one for in the
middle. And the RDM to the right. In the upper row is the
whole set and in the lower row (scaled to be larger) is the
set with five remaining outliers that gives a clear trace in
the matrices.
The homography can be computed using this set
of inliers and the set of inliers can easily be pruned
using ordinary RANSAC by setting a desired
threshold. As all matches are regarded as inliers
RANSAC will now without problem find all inliers
within that specific threshold.
4 CONCLUSIONS
The PUtative Match Analysis (PUMA) is a
repeatable, brute force algorithm that can be used
whenever it is necessary to test other parameters in
an application and when it is crucial that the set of
inliers are always the same for the same set of data.
Hence, it can be used as a robust tool for testing and
development of computer vision applications with
small non affine distortions as in the aerial images
used as examples. Due to its high computational cost
it will be outperformed by RANSAC in most
situations, but when there are many outliers, PUMA
can be a good choice for a reasonable number of
matches as it has proven to find the inliers even for a
rather high amount of noise in the set.
ACKNOWLEDGEMENTS
This work was carried out during the tenure of an ERCIM
"Alain Bensoussan" Fellowship Programme at IIT,
CNR in Pisa, Italy.
REFERENCES
Capoyleas, V., Rote, G. and Woeginger, G., 1991.
Geometric Clusterings, J. Algorithms, vol. 12, pp. 341-
356.
Chum, O., Matas, J., Kittler, J., 2003. Locally Optimized
RANSAC. In DAGM-Symposium. 236-243.
Chum, O., Matas, J and Obdrzalek, S., 2004. Enhancing
RANSAC by generalized model optimization. In
Proceedings of the Asian Conference on Computer
Vision (ACCV).
Chum, O. and Matas, J., 2005. Matching with PROSAC -
Progressive Sample Consensus. In Proceedings of the
IEEE Computer Society Conference on Computer
Vision and Pattern Recognition (CVPR). pp 220-226.
Chum, O. and Matas, J., 2008. Optimal randomized
ransac. IEEE Transactions on Pattern Analysis and
Machine Intelligence, 30(8). pp1472–1482.
Fischler, M. A. and Bolles, R. C., 1981. Random sample
consensus: A paradigm for model fitting with
applications to image analysis and automated
cartography, Communications of the ACM, 24, pp.
381–395.
Harris, C., Stephens, C., 1988. A Combined corner and
edge detection. In Proc. of The Fourth Alvey Vision
Conference, pp. 147–151.
Jain, A. K. and Dubes, R. C., 1988. Algorithms for
Clustering Data. Englewood Cliffs, N.J.: Prentice
Hall.
Lowe, D. G., 2004. Distinctive Image Features from
Scale-Invariant Keypoints, International Journal of
Computer Vision, 60, 2, pp. 91-110.
MacQueen, J. B., 1967. Some Methods for classification
and Analysis of Multivariate Observations,
Proceedings of 5-th Berkeley Symposium on
Mathematical Statistics and Probability, Berkeley,
University of California Press, vol 1, pp. 281-297.
Michaelsen, E., von Hansen, W., Kirchhof, M., Meidow,
J., Stilla, U., 2006. Estimating the Essential Matrix:
GOODSAC versus RANSAC, PCV06, pp.1-6.
Sattler, T., Leibe, B., Kobbelt, L., 2009. SCRAMSAC:
Improving RANSAC's efficiency with a spatial
consistency filter. ICCV 2009: pp. 2090-2097.
Vincent, E. and Laganiere, R., 2001. Detecting planar
homographies in an image pair. Image and Signal
Processing and Analysis, pp. 182–187.
Zuliani, M., 2009. RANSAC for dummies. pp. 42.
http://vision.ece.ucsb.edu/~zuliani/Research/RANSAC
/docs/RANSAC4Dummies.pdf
VISAPP 2012 - International Conference on Computer Vision Theory and Applications
344