SPOT 
ge 
ofined. 
| the 
age. 
3ses 
points 
nage 
nts 
or part 
> 
t 
ill 
atures 
it is 
of 
| of 
or 
The procedure is also pyramidal. Matching is 
performed for different spatial resolutions of 
the imagery. The images are subsampied and 
matching is performed. The resulting 
transformation is used to kick off the matching 
procedure for images sub sampled at a different 
level. 
2. SYSTEM DETAILS 
2.1 The System Components 
The system is designed in a modular manner. 
The system components are: 
* initial image registration 
* image smoothing 
* image subsampling 
e areas of interest 
e interest point extraction 
* feature point matching 
e intensity matching 
2.2 Initial Image Registration 
The procedure requires an initial image 
registration, to kick off the hierarchic and 
iterative matching procedure. The initial 
registration is defined in one of three manners. 
The user can define two tie points manually 
from which a Helmert transformation is defined 
using a method outlined by Maling (1991). 
Alternatively a single tie point is manually 
defined by the operator. This allows for the 
translational component of the transformation 
between the two images. The scene heading 
information from the SPOT image header is 
used to determine the rotational component of 
the transformation. 
An initial transformation can be determined 
solely from the SPOT image header 
information, using the latitude and longitude 
specified for the corner points and centre point 
of each scene. This is for SPOT products of 1A 
level processing. The initial transformation 
defined in this manner will normally be of 
poorer quality than that defined with control 
points. 
2.3 Image Smoothing 
A low pass Gaussian filter is applied to all 
imagery before features are extracted from 
the imagery. The width of the Gaussian 
function can be modified, resulting in variable 
degrees of smoothing. This procedure attempts 
to reduce high frequency noise in the imagery, 
and improve the results of the feature point 
extraction procedure. 
2.4 Image Subsampling 
The SPOT image is sub sampled to different 
resolutions. In practice this involves the 
selection of the nth pixel in each row and 
column. If this procedure is applied to a SPOT 
image pair for various resolution reductions 
then an image pyramid can be built up. The 
bottom of the pyramid is the full resolution 
image, the top of the pyramid is the image pairs 
with the most severe resolution reduction. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996