tanbul 2004
ze — Eine
ALGORITHMS AND EXPERIMENT ON SAR IMAGE ORTHORECTIFICATION BASED
ON POLYNOMIAL RECTIFICATION AND HEIGHT DISPLACEMENT CORRECTION
l'raffic Flow
Conference. G.M. Hume *, LK. Guo? 4.G. Lv", Z. Xiao Z. Zhao 8 CP. Qiu^
a Chinese Academy of Surveying and Mapping, Beitaiping Road, No.16, HaiDian, Beijing, 100039, China-
huang.guoman@163.net, littleboatxz@hotmail.com, zhengzhaochina@ 163.com, qep78@163.net
b Institute of Geographical Science and Natural Resources Research, CAS, 11A Datun Road, Anwai, Beijing, 100101, China-
Guojk@lreis.cn
znition from
nmetry and
62-166, CD
ni Suites KEY WORDS: SAR, Image, Orthorectification, Algorithms, Experiment, Accuracy
Systems, J
ABSTRACT:
This paper introduces the algorithm on SAR image orthorectification based on polynomial rectification and height displacement
correction, and experiments according to the algorithm on ERS-2, RADARSAT, and airborne SAR image in mountain area. Many
factors result in SAR image distortion, and most of them can be corrected by polynomial rectification. But the distortion of SAR image
brought by elevation is very difficult to be corrected by polynomial rectification. In this paper, height displacement caused by
elevation is corrected in advance according to the slant distance and elevation of each pixel, and the other distortion is corrected by
polynomial rectification. Compared to other orthorectification algorithm on SAR image, it is very easy to implement. According to
this method, programme has been designed. And then, experiment has been down on RADARSAT image in a mountain area in China
(Dali, Yunnan), and polynomial rectification on the same image also has been down by the same ground control points. The accuracy
of the former is about 2.8 pixels, and the latter 44.4 pixels. In this area, the most elevation difference of the ground control points is
about 2972 meters. Some other experiments have been down on ERS-2 image (Chengdu, Sichuan), and 3 meters resolution airborne
SAR image (Dabieshan, Anhui), and the results are similar to that of the first experiment. So, the new algorithm on SAR image
orthorectification introduced in this paper is efficient and practicable for SAR image orthorectification in mountain area.
1. INTRODUCTION
The geometric correction process seems to be more important
today than before (Thierry, 2003). Synthetic aperture radar
(SAR) adopts side-looking imaging mode, and the side-looking
angle of SAR image is much larger than that of optical image.
This mode leads to a great influence to geometric distortion of
SAR image. Consequently, it is very important for SAR
application to rectify geometric distortion and create
ortho-image.
Many SAR image rectification methods are put forward such as
orthorectification of Radarsat fine using orbital parameters and
DEM (Keong, 1995; Mohd, 2000), practical SAR
orthorectification (Leland, 1996) and geometry processing of
remote sensing images (Thierry, 2003). Now, the primary
methods of SAR image rectification include polynomial
rectification, collinearity equation method and the
Range-Doppler method.
Polynomial rectification
Since 1970s, polynomial functions are well known (Wong,
1975; Billingsley, 1983). Based on polynomial function,
polynomial rectification is a comparatively traditional method
for rectification, which is often applied to optical image
orthographic rectification. For SAR image of plant areas, it also
can be used to rectify geometric distortion.
Collinearity equation method
This method includes two types. The first one is mathematical
model, which presenters are F.Leberl etc. Change of linear
elements in sensor’s exterior azimuthal elements are considered,
nor do angle elements. Therefore, after establishing SAR stereo
model, there are biggish fluctuate parallax, and it is only the
same with airborne SAR because this model is building in terms
with the range equation of image points and zero Doppler
condition. The other one is mathematical model of flat range
projective radar images performed by G.Konecny etc. In this
model, the changes of exterior azimuthal elements of sensors
and terrain are considered, and the form of equation is similar to
photogrammetric collinearity equation. Although the model is
casy to be applied, it interpret something only referring to the
characteristic of traditional optical imaging without taking into
account SAR image side-looking projective characteristic.
Forasmuch, this type model is only a simulant processing
method to optical image.
The Range-Doppler method
The Range-Doppler method primarily discusses the relationship
of image points and target points {rom the view of SAR imaging
geometry. The following is its basic theory: In the range
direction, the distribution of equidistance points from target to
radar is homocentric circular cluster, which center of a circle is
point bellow satellite. In the other hand, equal Doppler
frequency shift points created by relative moving between
satellite and target distribute as hyperbolic cluster in the
azimuth direction. Therefore, ground target can be confirmed by
the intersection of the clusters of homocentric rotundity and
hyperbola. The Range-Doppler algorithm mostly lies on the
accuracy of fundamental catalogue data.
Polynomial rectification regards general distortion of remote
sensing image as combination of several of basic and high
distortion. For relatively flat areas, it can reach sufficient
rectification accuracy and is easy to use. Accordingly, it has
been used in considerable fields. But considering some more
hypsography areas, this method can not lead to satisfying results,
especially in the condition of biggish slope. In this paper, we
add this method to correction of height displacement caused by
elevation, and it can improve corrective accuracy. At first,
height displacement is corrected; and then the other distortion is
corrected by common polynomial rectification. By contraries
during resampling, we firstly extract the coordinates of image
points which haven't been affected by elevation in terms of
139
