Page 225 - Digital Analysis of Remotely Sensed Imagery

Image Geometric Rectification 189

except that they should be representative of the whole planimetric
and height range with a balanced distribution. This means that GCPs
should be dispersed throughout the area covered, and TPs should
be located widely within the overlapped images. Unlike GCPs that
must be identifiable on the ground or on topographic maps, TPs need
to be identifiable only in the overlapped portion of adjoining images
or photographs. Some of the rectification models introduced in Sec. 5.4
are also useable for image orthorectification, including the simple affine
transformation and projective transformation models. In particular,
the traditional differential rectification model is able to remove topo-
graphic relief displacement. In addition, several sensor-specific mod-
els such as RPCs are also applicable to orthorectification. At present
the RPC method applies to IKONOS Geo Ortho Kit images and
QuickBird Other Ready Standard imagery, OrbView-3, and SPOT.
Supplied with these images is an ancillary file containing the RPC
parameters. With the increasing coupling of sensors with GPS units,
it is possible for the RPC model to find more applications to other
very high resolution satellite images in the near future.
A critical preliminary step in image orthorectification is the
construction of a DEM. It should cover the same geographic area
as the imagery being rectified. Preferably, it should have the same
spatial resolution as the imagery. For instance, the DEM should
have a cell size of 30 m if the imagery being orthorectified is from
Landsat TM. DEMs can originate from a wide range of sources.
Coarse resolution (e.g., 90 m) DEMs, such as those obtained from
the Shuttle Radar Topography Mission, are freely available for the
global terrestrial surface. DEMs at a spatial resolution finer than
this have to be constructed from stereoscopic pairs of aerial photo-
graphs by means of digital photogrammetry. Alternatively, they
can be created from existing digital contour data, TIN, or raster
data via a process known as spatial interpolation (Fig. 5.21). With
increasing ease of availability, light detection and ranging (LiDAR)
data are another source of reliable elevational information. It is
important to construct the DEM at the highest accuracy possible
because the horizontal shift in the position of pixels is calculated
from its elevation in the DEM. The accuracy of elevation informa-
tion governs the quality of image orthorectification.
Prior to the application of the constructed DEM in orthorectifica-
tion, attention must be paid to the datum from which elevation is
referenced, in particular, whether the DEM and the height of the sen-
sor are referenced to the same datum. Usually called the optometric
height, elevations in most DEMs are referenced to the mean sea level.
Globally, mean sea level is a broadly undulating surface known as the
geoid (Fig. 5.22). In contrast, satellite height is referenced to an ideal
earth-centered ellipsoid whose geometric shape is mathematically
defined, such as the World Geodetic System (WGS) 1984. The vertical
discrepancy between the geoid and ellipsoid surface at any location,