OPTIMIZED PATCH BACKPROJECTION IN ORTHORECTIFICATION FOR HIGH 
RESOLUTION SATELLITE IMAGES 
Liang-Chien Chen *, Tee-Ann Teo, Jiann-Yeou Rau 
Center for Space and Remote Sensing Research, National Central University, Chung-Li, Taiwan. 
(Icchen,ann,jyrau)@csrsr.ncu.edu.tw 
Commission ICWG IVIV 
KEY WORDS: Orthoimage, High Resolution, Satellite, Orientation, DEM/DTM 
ABSTRACT: 
The objective of this investigation is to build up a fast orthorectification procedure for high resolution satellite images. The 
proposed scheme comprises two major components: (1) orbit modeling, and (2) image orthorectification. In the orbit modeling, we 
provide a collocation procedure to determine the precision orbits. In the image orthorectification, the area of interest is sequentially 
subdivided into four quadrate tiles until a threshold is met. The threshold of maximum terrain variation in a tile will be optimized 
according to the computation efficiency and accuracy requirements. Once the ground tiles are determined, we perform adaptive 
patch backprojection to correspond to the image pixels. Selecting the highest elevation in the tile, the four corners of the tile are 
projected on the image to form a set of anchor points. Another set of anchor points with the lowest elevation are generated in the 
same manner. Assuming that the relief displacement in a moderate tile is linear, a groundel within the tile is projected into the 
image space according to the groundel elevation and the two associated anchor point sets. Tests of images include SPOT 5 
supermode and QuickBird panchromatic satellites. Experimental results indicate that the computation time is significantly reduced 
without losing accuracy. 
1. INTORDUCTION accuracy. Inspired by the idea, we propose a “Patch 
Backprojection” procedure for accelerating the computation in 
The most rigorous way to register a remotely sensed image with ^ orthorectification for high resolution satellite images with large 
a relevant spatial data layer is performing orthorectifictation to amount of pixels. 
the image. The generation of orthoimages from remote sensing 
images is an important task for various remote sensing Because of the small field-of-view (FOV) of high resolution 
applications, such as cartography, environmental monitoring, satellite, the relief displacements in a small area with moderate 
city planning, etc. Moreover, GIS (Geographic Information terrain variations may be assumed linear. We, thus, propose a 
Systems) technology often needs multi-temporal images for method to do the orthorectification patch by patch. The patch 
detection of lancover changes. Thus, ortho-rectified images size may be adapted for different terrain characteristics. We 
have become important due to their short production time. first divide the area of interest into a number of tiles. For 
corner point with highest elevation, we compute the image 
Nowadays, most of the high resolution satellites use linear coordinate for each corner point of tiles using indirect method. 
pushbroom arrays, such as SPOTS, lkonos, QuickBird and The indirect method also applies to the corner point with lowest 
others. A number of investigations have been reported elevation. Using an affine transformation as a mapping 
regarding the geometric accuracy for those pushbroom linear function of image coordinates and object coordinates. In 
array images (Westin, 1990; Chen and Lee, 1993; Orun and addition, we will analyze terrain variations for the selection of 
Natarajan, 1994; Toutin, 2003). Traditionally, the first step of the adaptive window of the tiles. We also analyze the model 
image orthorectification is to model the orientation parameters error of the proposed method including transformation error and 
by using ground control points. Then, incorporating a DTM, an interpolation error. Affine transformation, patch size, tilt angle, 
image, and the orientation parameters, a non-linear equation is and elevation range are the most important factors to be 
formulated to determine the along-track image coordinates in considered. 
terms of the sampling time for a ground element. The across- 
track image coordinates can thus be calculated according to the In the validation, we first analyze the model error of the 
collinearity condition equations. proposed method. It has two parts, transformation error and 
interpolation error. Affine transformation, patch size, tilt angle, 
The traditional solution of orthorectification for pushbroom and elevation range are the most important factors to be 
images is time-consuming due to a vast amount of non-linear considered in the analysis of model errors. Then, we check the 
equations have to be solved. This weakness is so obvious for accuracy of the determined orientation parameters. Finally, the 
those high resolution satellite images that an efficient way is accuracy of the generated orthoimage will be examined. 
required.  Konecny et al (1987) emulated SPOT images as Pushbroom scanner images including SPOTS and QuickBird 
centre perspective, then, implemented the idea on an analytical are considered in this investigation. 
plotter to achieve real time operation while maintaining some 
  
* Corresponding author. 
586 
Intern 
The p 
is to 
contre 
perfor 
Back 
“Patcl 
2] « 
In the 
and tl 
polyn: 
high « 
attitud 
a coll 
errors 
The fi 
board 
low o 
Once 
detern 
Least 
1982). 
22 I 
The o 
corres, 
provid 
a “Pat 
approe 
perfor 
efficie 
metho 
grounc 
2.2.1 
Figure 
grounc 
to ime 
princip 
plane. 
to the 
the inr 
define 
f(t)=1 
We ar 
equatic 
point 
coordir 
princip 
point f 
calcula 
by pixe