ADVANCEMENT IN MATCHING OF SPOT IMAGES BY 
INTEGRATION OF SENSOR GEOMETRY AND TREATMENT OF 
RADIOMETRIC DIFFERENCES 
Emmanuel P. Baltsavias, Dirk Stallmann 
Institute of Geodesy and Photogrammetry 
Swiss Federal Institute of Technology 
ETH-Hoenggerberg, CH-8093 Zurich, Switzerland 
Tel.: +41-1-3773042, Fax: +41-1-3720438, e-mail: manos@p.igp.ethz.ch text 
Commission IV m 
VIE) 
ABSTRACT acti 
COV! 
This paper presents a matching algorithm for automatic DTM generation from SPOT images that provides dense, met 
accurate and reliable results and attacks the problem of radiometric differences between the images. The edgi 
proposed algorithm is based on a modified version of the Multiphoto Geometrically Constrained Matching 
(MPGC). It is the first algorithm that explicitly uses the SPOT geometry in matching, restricting thus the search 
space in one dimension, and simultaneously providing pixel and object coordinates. This leads to an increase in 
reliability, and to reduction and easier detection of blunders. The sensor modelling is based on Kratky's AS 
polynomial mapping functions to transform between the image spaces of stereopairs. With their help epipolar Swi 
lines that are practically straight can be determined and the search is constrained along these lines. The opti 
polynomial functions can also provide approximate values, which are further refined by the use of an image to a 
pyramid. a 
i 
Radiometric differences are strongly reduced by performing matching not in the grey level but in gradient agri 
magnitude images. Thus, practically only the information in stripes along the edges is used for matching. Edges part 
that exist in only one image can be detected by subtracting quasi registered images in the upper levels of an ran 
image pyramid and by consistency checks. The points to be matched are selected by an interest operator in was 
preprocessed gradient images. Gross errors can be detected by statistical analysis of criteria that are provided by 
the algorithm. ar 
The results of an extensive test using a stereo SPOT model over Switzerland will be reported. Different cases of AM 
radiometric differences, and matching with different options and the qualitative comparison of the results based 136 
on forty thousand check points will be presented. rigc 
KEY WORDS: SPOT, automatic DEM generation, image matching, geometric constraints, radiometric se 
differences, accuracy analysis me: 
SEC 
1. INTRODUCTION approximation a straight line. If a straight line is defined poii 
); : : : by the projection of a small ray segment which is centred 
The motivation behind this research was the aim to at the ren point position in ver space, then the 
improve matching of SPOT images by integr ation of the deviation of the epipolar line from the straight line would 
Sensor geometry and tec treatment of radiometric be 0.25 pixel for a height error of more than 7 km. More MF 
itcrenices: AS lar ds foe authors know, none of the details on the characteristics of the PMFs can be found in cor 
published matching methods explicitly exploits the SPOT Baltsavias and Stalimann. 1902 
: : ; : get 
geometry to restrict the search space. Since SPOT does cor 
not have a perspective geometry in flight direction, Attempts to circumvent the problem of radiometric in ; 
accurate epipolar images can not be generated without differences have concentrated on using images acquired $6 
the use of a DTM and thus the existing matching within a short time interval. However, this is difficult to Tal 
algorithms perform a 2-D search. The authors achieve and does not solve the problem. The along-track det 
investigated Kratky’s polynomial mapping functions stereo will strongly reduce these problems but can not nu 
(PMFs) (Kratky, 1989a) which transform from image to eliminate them due to different perspective views, clouds sin 
image, image to object and object to image space. For the ‘and occlusions. Fusion and matching of multitemporal alo 
computation of the coefficients of the PMFs the results of ~~ and multisensor data, as in change detection applications, 
a rigorous bundle adjustment are used. The PMFs are will retain their importance even in the era of along-track ge 
much faster and almost equally accurate as the rigorous stereo. Thus, the authors decided to attack this problem, Cr 
transformations. When the ray of an imaged point is which has been up to now treated only to a limited extent. 
projected by using the PMFs onto the other image of a The idea is to use gradient magnitude images, thus 
stereo pair (epipolar line) it is with a very good eliminating radiometric differences in areas of low 
916