2.7.6 Tile Matching 
Feature matching is generally not performed 
over the whole image, but rather for tiles 
within the image. Tiles are used because a low 
order transformation cannot be extended over a 
whole SPOT scene without the introduction of 
unacceptable errors, during the matching 
process, 
Large images are split into tiles of 
approximately 500 by 500 pixels. Matching 
will occur within these tiles. In practice only 
the first image of the image pair is tiled 
because points within a tile in the first image 
can be matched to points in several tiles in the 
second image. 
A geometric transformation is calculated for 
each tile, and these must be appropriately 
propagated from one level of the pyramid to the 
next. 
When matching has been completed for a tile, a 
tile from the first image is cross correlated 
with its corresponding area in the second 
image, based on the most recently defined 
image transformation. Again resampling will 
occur if rotation or scale differences require 
this. If the cross correlation is less than an 
operator specified threshold, then the 
transformation for the tile is rejected. This 
tile is then frozen and no more processing is 
performed on the tile at any level of the 
pyramid. 
2.8 Image Based Intensity Matching 
Intensity matching is performed only at the 
bottom of the pyramid, rather than at every 
level of the pyramid. Intensity matching is a 
final check on the validity of each conjugate 
point pair, either accepting or rejecting the 
pair, based on considering the grey level values 
of pixels around each point of the conjugate 
point pair (Ackermann 1984). 
Given a conjugate point pair an image chip is 
considered around the points in each image. 
These two chips may differ due to radiometric 
and geometric inconsistencies. A 
transformation is sought to take into account 
these elements. Starting with approximate 
values for the unknown radiometric and 
geometric coefficients a least squares solution 
is found for the transformation between the 
two windows. The initial coefficients are 
updated after the first iteration and the second 
image chip is resampled based on the geometric 
coefficients. The procedure is iterated till 
convergence is achieved. If convergence is not 
achieved after some iterations then the 
conjugate point pair is rejected. 
At the end of this procedure a trimmed down 
set of conjugate points can be XPECTed. 
2.9 Transformation Calculation and 
Resampling 
Procedures exist to allow the calculation of 
geometric transformations from a set of tie 
points. These are usually applied after the 
intensity matching. Based on a geometric 
transformation a SPOT image can be resampled, 
affecting the co-registration of two images. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996 
in à