

Title of Thesis
Level3 Geometric Correction of FORMOSAT2 Imagery and Efficient
Image Resampling 
Author(s)
Faheem Arif 
Institute/University/Department
Details Department of Computer Science / National
University of Science & Technology, Rawalpindi 
Session 2009 
Subject Computer Science 
Number of Pages 124 
Keywords (Extracted from title, table of contents and
abstract of thesis) Level3, Geometric, Correction,
distortions, FORMOSAT2, satellite, Imagery, Efficient,
Image, Resampling, wavelettransform 
Abstract A significant
problem in satellite imagery is geometric distortion. Accurate
remote sensing and high resolution satellite images have made it
necessary to revise the geometric correction techniques used for
orthorectification. Conventional methods of photogrammetric
modeling of remotely sensed images are insufficient for mapping
purposes and need to be substituted with more rigorous approach to
get a true orthophoto.
FORMOSAT2, a newly launched remote sensing Taiwanese satellite, has
high spatial resolution sensor onboard for a daily revisit orbit.
However, like any image acquisition system, it also produces
geometric distortions in its raw images. Pixel Projection Model
(PPM) was devised by National Space Program Office (NSPO) Taiwan,
for processing of Level1A (Raw) satellite images to Level2 (radio
metrically corrected) images. Being systematically corrected,
Level2 images still possess terrain elevation, rotationtranslation
and geometric distortions. There was a dire need for enhancement of
this model to produce Level3 (geometrically corrected) image
products.
A novel method for Level3 correction of satellite images,
especially suited for FORMOSAT2, has been developed. The PPM has
been enhanced to cater for geometric distortions caused by the
attitude change in the satellite specifically in the preprocessing
stage. The three attitude angles of the satellite are thus
calculated and corrected as per the ground position or coordinates
using least squares adjustments. The approach is based on
nonsystematic method in which physical modeling of the satellite
imagery is considered. The mathematical model has been developed to
calculate and correct instrument bias/ attitude angles. Ground
Control Points have been integrated in the algorithm besides vertex
matching for more precise results. Results were verified by
computing MSE for image to image matching and point to point
matching. An improvement of 86.3% was obtained for the new Level3
correction technique over the existing Level2 algorithm.
Three conventional interpolation techniques for transformation of
image pixels to earth coordinate system were also analyzed for
improvement. The experimental results show that the cubic
convolution based modeling is best suited for output pixel value
transformation but it is computationally complex with a higher
execution time. To improve this, a wavelettransform based filter (Daubechies
4) was developed for image pixel transformation. The new method
provides similar visual interpretation as cubic convolution but with
much lower computational complexity and execution time. The proposed
wavelettransform based method is an order of magnitude faster than
the cubic interpolation technique.
Level3 geometrically corrected FORMOSAT2 images can be used for
disaster investigation/ prediction, environmental monitoring,
vegetation evaluation, and multitemporal image matching. In our
work, we have focused on the application of geometrically corrected
imagery for disaster investigation.

