International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
GEOCODING AND COREGISTRATION OF MULTISENSOR AND MULTITEMPORAL
REMOTE SENSING IMAGES
Hannes Raggam, Mathias Schardt and Heinz Gallaun
Institute of Digital Image Processing, Joanneum Research, Wastiangasse 6, A-8010 Graz, Austria
{hannes.raggam} {mathias. schardt} {heinz.gallaun} @joanneum.ac. at
KEYWORDS: Geocoding, Parametric Imaging Models, Coregistration, Image Matching, Polynomial Rectification.
ABSTRACT
The analysis of multitemporal/multisensor remote sensing datasets can only be efficiently done if the data refer to a common
geometry. Geocoding of the individual images of such a dataset to the geometry of a topographic map is the most usual procedure to
accomplish data comparability. Registration of the images will have to meet extremely strict requirements, if data acquired at
different dates and/or with different systems will be processed simultaneously in one "data stack". The approaches to precisely
coregister data from a geometric point of view can be based on (a) the utilisation of sensor specific parametric imaging models to
precisely relate image pixels to ground, or (b) the utilisation of general coregistration techniques, which are independent of any
sensor-specific imaging parameters. In general, parametric approaches are increasingly used for geocoding. The accuracy achieved
with such methods lies in the order of one pixel. However, the experience from various applications, e.g. change detection analysis,
showed the necessity of achieving mean geometric accuracies of less than half a pixel, if pixel-by-pixel comparison is to be
performed. This is particularly true for classes, which are characterised by heterogeneous spatial distribution such as mixed forests,
small patterns of agricultural areas or settlements. This required accuracy can be obtained under certain conditions by automated
image matching and coregistration of the images with acceptable efficiency. The paper will discuss the performance of both
parametric methods and matching-based coregistration techniques with regard to monitoring applications that are currently
developed at the Institute of Digital Image Processing, Joanneum Research.
1. INTRODUCTION
The integration of multitemporal and multisensor remote
sensing data and other relevant data layers demand appropriate
data coregistration methods. Only if such methods are available,
the new information resulting from the combination of the
source datasets can be optimally utilised by the various users.
By applying data coregistration methods, new information can
be gained from the complementary information content of the
multiple data sources.
According to various authors, data fusion can be done at
different levels. Pohl and van Genderen (1998) distinguish
between data fusion at pixel, feature or decision level. Csatho
and Schenk (1998) use a slightly different classification,
considering fusion at signal, pixel, feature and symbol level. In
either case, it is recognised that the data have to undergo a
respective pre-processing, including geometric but also
radiometric procedures. These pre-processing steps are in
general a very critical point for operational data fusion
applications and, therefore, can not be seen independently from
the applied data fusion methods.
This paper is related to the geometric pre-processing of
multitemporal and multisensor images, so that a subsequent
pixel-level image fusion is possible. This geometric pre
processing comprises registration or geocoding of the images,
such that their corresponding pixels have the same geometry,
i.e. the aim is to generate precisely coregistered datasets (see
Figure 1).
In this paper, the following approaches to generate coregistered
images in either image or map 1 geometry will be generally
discussed:
• Polynomial rectification
• Parametric geocoding
• Matching-Based Coregistration (MBC)
The discussion on MBC will focus on approaches as
implemented and used at the Institute of Digital Image
Processing. In this context, polynomial rectification methods are
1 Here, map stands for an object space reference coordinate
system.
