×

You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Title
Fusion of sensor data, knowledge sources and algorithms for extraction and classification of topographic objects
Author
Baltsavias, Emmanuel P.

International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
many years and many applications. However, polynomial
transformations should not be taken into consideration for
precise image registration. In some cases, polynomial
transformations can be used, e.g. an affine one could be used to
transform an already geocoded image from one UTM zone to an
adjacent one, or to register multi temporal images acquired by
the same sensor under very similar imaging conditions, like it
happens for Landsat TM. They can also be used, if the terrain is
not rough and can be approximated by an appropriate
polynomial.
Polynomial and image warping methods are based on the use of
2-D tie-points between datasets to be registered. Image
distortions caused by the terrain topography are usually not
taken into account. Following for instance a geometric
interpretation of a polynomial-based rectification, an image is
mapped onto an artificial “tie-point”-surface defined by the
points being used. This may cause severe displacements in
comparison to a precise orthorectification of the image onto the
real terrain or the reference ellipsoid surface, respectively (see
Figure 2).
3.2, Parametric Methods
Parametric techniques are based on the use of sensor-specific
mapping models in order to relate image and map data. For
representation of the terrain surface, a digital elevation model
(DEM) is usually incorporated into these methods. These are
the essential prerequisites to establish a precise relation between
a point on ground and the related pixel in an image (Figure 3).
Images can then be registered either in image or map geometry.
Registration of images in map geometry is in general achieved
through geocoding to a common geometry referring to a defined
map projection. Registration in image geometry is achieved
through a combination of geocoding (image-to-map) and map-
to-image registration techniques. Geocoding is the standard
approach, as the output can be immediately compared to other
geocoded products.
Fig. 3. Scheme of parametric relation between image and real
topography, represented by a DEM.
3.2.1. Geometric Modelling
Parametric methods rely on the availability of a sensor-specific
imaging model. The establishment of models to map image to
object space and vice versa is called geometric modelling in our
terminology and is the essential preparatory step for further
registration or geocoding.
In the geometric modelling task, any relevant sensor-specific
parameters must be provided in order to establish a precise
relation between an image pixel and the corresponding ground
location. Such parameters include both information on the
sensor geometry (e.g. pixel size, focal length etc.) as well as the
position and attitude of the sensor during image acquisition.
Latter information is delivered as ancillary information together
with the image data for the majority of current remote sensing
sensors. The information on sensor geometry, position and
attitude is used to derive initial approximations for the
parameters of the geometric model (initial imaging model).
However, the geometric parameters being provided are
generally not accurate enough, or part of the relevant imaging
parameters is even missing. Ground control points (GCPs)
measured in the image and the reference coordinate system are
therefore usually required in order to derive approximations for
the missing geometric imaging parameters and to accurately
estimate the entire set of imaging parameters through the
application of least squares adjustment procedures. This leads to
an optimised imaging model (see Figure 4).
The accuracy resulting from optimised imaging models is
mainly driven by the localisation accuracy of the GCPs in the
image, which depends on the ability to exactly identify the