In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
The type of basis function, g, determines the influence of each 
CP on the RBF, that is, the CP scope. So, the accuracy of the 
registration depends extremely on the distribution of CPs on the 
image. In this work we employ the thin plate spline (TPS) fun 
ction g(rj) = r 2 log r 2 (Bookstein, 1989), which is perhaps the 
RBF most widely employed for elastic registration. 
To evaluate the method performance, we have compared the re 
gistration accuracy obtained using the resultant CP set with res 
pect to uniform and random CP distributions. The uniform distri 
bution is obtained by selecting CPs according to a regular grid of 
squared cells of 50 pixels of side, while the random distribution 
is obtained by arbitrarily selecting the same number of CPs than 
the uniform one. 
The results of the comparison, displayed in figure 3, show how 
the proposed method yields better results in terms of accuracy. 
The accuracy of the registration process has been assessed com 
paring the geometric errors (RMSE and CE90%) of a set of inde 
pendent control points (ICPs) manually identified, achieving on 
average RMS errors under 1.4 m. with CPs distributed according 
to the DTM information. 
RMSE (m.) 
Figure 3: Accuracy of the proposed method compared to uni 
form and random distribution of CP considering a) RMSE and b) 
CE90%. 
Observe that the results of the uniform distribution and our ap 
proach are similar, since the smallest squared cell generated by 
our approach has the same size that the one considered in the uni 
form distribution. The number of CP required in our approach, 
however, is, on average, around 37% lower. The benefits of our 
approach are clear, specially, when the CP extraction must be ma 
nually performed. 
4. CONCLUSIONS 
This paper presents a technique to distribute the CP pairs accor 
ding the relative image distortions, more severe in rugged te 
rrains, and proposes an automatic procedure to robustly extract 
CPs in two images by applying computer vision techniques. The 
experimental results reveal the advantage of employing our met 
hod, in comparison with other two strategies (uniform and ran 
dom distributions) implemented in most of popular commercial 
packages of remote sensing like ERDAS, ENVI and PCI. 
REFERENCES 
Arevalo, V. and Gonzalez, J., 2008. An Experimental Evalua 
tion of Non-Rigid Registration Techniques on QuickBird Satellite 
Imagery. International Journal of Remote Sensing 29(2), pp. 513— 
527. 
Bookstein, F., 1989. Principal Warps: Thin-Plate-Splines and the 
Decomposition of Deformations. IEEE Transactions on Pattern 
Analysis and Machine Intelligence 11(6), pp. 567-585. 
Burt, P. and Adelson, E., 1983. The Laplacian Pyramid as a Com 
pact Image Code. Communications, IEEE Transactions on [le 
gacy, pre-1988] 31(4), pp. 532-540. 
Fischler, M. and Bolles, R., 1981. Random Sample Consensus: A 
Paradigm for Model Fitting with Application to Image Analysis 
and Automated Cartography. Communications of the ACM 24, 
pp.381-395. 
Fonseca, L. M. G. and Kenney, C. S., 1999. Control Point Assess 
ment for Image Registration. In: Proc. XII Brazilian Symposium 
on Computer Graphics and Image Processing, pp. 125-132. 
Elarris, C. and Stephens, M., 1988. A Combined Comer and Edge 
Detector. In: Alvey Vision Conference, Vol. 4, Manchester, UK, 
pp. 147-151. 
Hartley, R. and Zisserman, A., 2004. Multiple View Geometry 
in Computer Vision. Second edn, Cambridge University Press, 
Cambridge, UK. 
Lucas, B. and Kanade, T., 1981. An Iterative Image Registration 
Technique with an Application to Stereo Vision. In: International 
Joint Conference on Artificial Intelligence, pp. 674-679. 
Schowengerdt, R., 2007. Remote sensing: Models and Methods 
for Image Processing. Academic Press. 
Torr, R, 2002. A Structure and Motion Toolkit in Matlab: Interac 
tive Adventures in S and M. Technical Report MSR-TR-2002-56, 
Microsoft Research. 
ACKNOWLEDGEMENTS 
DigitalGlobe QuickBird imagery used in this study is distribu 
ted by Eurimage, SpA. (http: //www. eurimage. com, accessed 
1 Jun. 2010) and provided by Decasat Ingenieria S.L., Malaga, 
Spain. 
This work has been partly supported by the Spanish Government 
under research contract CICYT DPI-2008-03527.