International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
Int( 
26 
3.3. Matching-Based Coregistration (MBC) 
3.3.1. Method 
Matching-based coregistration is a general technique to match 
geometrically images or images and maps, independently of 
their imaging geometry or projection in case of geocoded 
datasets. An automatic image matching procedure is applied 
(pixel-by-pixel or at coarser intervals) between an input image 
and a reference image in order to find corresponding points in 
the two images. Therefore, automated image matching is the key 
tool for this type of coregistration. The scheme of MBC is 
shown in Figure 7 for an input image in its original geometry 
and a reference image given in map/ground geometry, e.g. an 
ortho-image. 
A general workflow of the MBC procedure is shown in Figure 
8. It comprises the following processing steps: 
1. An automatic image matching procedure is applied to the 
image pair to be registered. The output of this procedure is a 
so-called disparity map, which shows the shift vectors 
between reference image and input image pixel co 
ordinates. In order to increase the performance of the image 
matching, the following pre-processing steps may be 
applied to the data: 
• radiometric equalisation in case that the images are 
radiometrically different, as is often the case for 
multisensor and multitemporal images; 
• coarse geometric registration through measurement of 
a few tie-points in case that the images are significantly 
different in geometry. 
Both steps can be done implicitly inside the matching 
procedure. 
2. In a second step the disparity vectors resulting from image 
matching are used to resample the input image to the 
reference image geometry. 
3.3.2. Application Areas of MBC 
Areas of application of the MBC technique include: 
1. Registration of multitemporal and multisensor images to 
the geometry of a selected reference image. 
2. Fine-registration of geocoded multitemporal and 
multisensor images in case that there are still obvious 
mismatches between the individual images after geocoding. 
In such cases, even small geolocation errors may disturb 
further applications. 
3. MBC may also be used to support fully automatic 
geocoding without any operators interaction. For that, an 
image may first be coarsely geocoded using an initial 
(coarse) parametric model (or even a polynomial 
rectification). In a second step, this coarse product may be 
geometrically refined using coregistration to an already 
geocoded image. 
Automatic image matching may also be applied in order to 
check the correspondence of any registered or geocoded images. 
The disparity vectors, which are achieved for such data will 
provide a neutral quality measure for the correspondence of the 
data, which obviously should be acceptably small. 
Grey value 
3.4. i 
Automat 
to the ne 
radiomei 
matching 
and used 
Researcf 
pixel m2 
moved c 
the resp 
differenc 
or featur 
The ma 
product 
matching 
is defin 
coefficie 
found in 
method, 
various 1 
image, ii 
the searc 
minimur 
correspo 
in Paa: 
backmat 
be applit 
3.5. i 
Methods 
Digital I 
multisen 
Buchroit 
also usei 
been de\ 
During t 
image g 
Remote 
Researcl 
geometr 
images. 
ERDAS, 
• Gee 
acq 
• Inte 
par; 
• Hig 
• Ste: 
ster 
• Gei 
froi 
• SA 
pha 
• DE