KEY TECHNIQUE OF ACCU 
RATE RECTIFICATION 
FOR REMOTE SENSING IMAGE WITH HIGH RESOLUTION 
Jianging Zhang, Yong Zhang*, Shunyi Zheng, Hongwei Zhang, Zhijiang Li 
School of Remote Sensing and Information Engineering, Wuhan University, Wuhan, China, 430079 - 
e 
(jqzhang,yongzhang)@supresoft.com.cn, syzheng(@263.net, Zhwlx_wuhan(@sina.com, zhijiang lee(yahoo.com.cn 
Commission III, WG 111/72 
KEY WORDS: Remote Sensing Image, High Resolution, Parameter Calculation, Image Matching, Accurate Rectification 
ABSTRACT: 
The geo-referenced image with one-meter resolution (GRIOMR) 
sensing image with | 
is one of the most basic information in the digital earth. The remote 
righ resolution (RSIHR) is the best resource for acquiring the GRIOMR in low costing and sort period. The key 
technique producing GRIOMR from RSIHR is introduced in this paper, including parameter computation of RSIHR, image 
matching to creating DTM and fast orthogonal rectifica 
tion of RSIHR. After the Rational Polynomial Coefficient (RPC) used by 
IKONOS and QUICKBIRD images and the block adjustment based on RPC parameters are presented, the new, simple and strict 
geometric model based on affine transformation is destribed. When there are control points less th 
an 5 in each image, the block 
adjustment based on RPC parameters could be used. If there are control points more than 4 in each image, the new, simple and strict 
geometric model based on affine transfor 
relative registration, remains quite large y-para 
from RSIHR. Based on DTM and image parameters acquired by block adjustment b 
Igorithm. The experimental results with real RSIHR show the strategy 
affine transformation, the RSIHR can be rectified by fast a 
mation should be applied. For some of RSIHR, the approximate epipolar image pair, after 
lax. Therefore, the 2 D relaxation matching should be used in the DTM modelling 
ased on RPC parameters or the model based on 
introduced in this paper is feasible. Corresponding conclusion and future work are summarized in the final. 
1. INTRODUCTION 
The geo-referenced image with one-meter resolution (GRIOMR) 
is one of the most basic information in the digital earth. The 
remote sensing image with high resolution (RSIHR) is the best 
resource for acquiring the GRIOMR in low costing and sort 
period. The key technique producing GRIOMR from RSIHR is 
introduced in this paper, including parameter computation of 
RSIHR, image matching to creating DTM and fast orthogonal 
rectification of RSIHR. 
Because of very strong relativity of traditional parameters of the 
RSIHR, traditional image parameters can't be acquired 
sometimes. The algorithm of fitting based on reasonable 
polynomials, proposed by Kratky (Kratky, 1989a and Kratky, 
1989b), is used. For example, the IKONOS and QUICKBIRD 
images are supported by RPC (Rational Polynomial Coefficient) 
(Z hang et al., 2001). But, the coordinate accuracy computed by 
them is quite lower in many cases, and some control points are 
still needed for improving the accuracy. Furthermore RPC/RPB 
model is too complex. After the block adjustment based on RPC 
parameters 15 presented, the new, simple and strict geometric 
model based on affine transformation is described (Zhang et al., 
2002). Only 8 affine coefficients and one slantwise angle need 
to be determined by control points. Then, the image coordinates 
can be calculated using ground coordinates and 9 parameters. 
When there are control points less than 5 in each image, the 
block adjustment based on RPC parameters, where an affine 
transform of image coordinates are computed by measured 
coordinates of control points and their coordinates calculated by 
RPC/RPB parameters, could be used. If there are control points 
KL 
* Corresponding author. 
more than 4 in each image, the new, simple and strict geometric 
model based on affine transformation should be applied. 
For some of RSIHR, the approximate epipolar image pair(Zhang 
et al., 1989), after relative registration, remains quite large y- 
parallax. Therefore, the usual 1-D image matching method is 
not suitable for the approximate epipolar image pair of the 
RSIHR, including the very efficient relaxation matching 
(Baltsavias, 1991; Zhang et al, 1992). Instead of the ! D 
relaxation matching, 2 D relaxation matching should be used in 
the DTM modeling from RSIHR. The procedure includes 
Feature point extraction, approximate value estimation, 
matching and refining and filter. 
Based on DTM and image parameters acquired by block 
adjustment based on RPC parameters or the model based on 
affine transformation, the image coordinates can be calculated 
for the orthogonal rectification of RSIHR. After re-sampling on 
original RSIHR, the orthogonal image is obtained. The fast 
algorithm for orthogonal rectification of RSIHR should be 
applied. The experimental results with real RSIHR show the 
strategy introduced in this paper is feasible. 
2. PARAMETER COMPUTATION OF RSIHR 
If there are control points less than 5 in each image, the block 
adjustment based on RPC parameters could be used. Otherwise 
the new strict geometric model based on affine transformation 
could be applied for any RSIHR. 
1096 
pai 
fol 
pai 
im: 
acc 
ima 
COO 
me: 
errc 
par: 
coo 
pare 
be c 
and 
2.2 
tran 
To 
RSI] 
affin 
the s 
y) ar 
2.2.1 
equat 
calcu 
simpl