Michael Breuer
4.1 Hyperspectral Image Data
The hyperspectral image data are of course the essential data in the following context. They need to be corrected. The
import of the hyperspectral data is mostly the starting point of the postprocessing pipeline. Today the data come on tape
or CD-ROM. The common formats are BSQ (band sequential) or BIL (band interleaved). Often auxiliary data such as
position and attitude measurements are stored together with the image data. This is the reason why it is difficult to
define standards for the storage of hyperspectral data because the kind of data often varies from one sensor to another.
Therefore a specific import routine is needed for an individual scanner type. In principle the individual format has to be
well described to be able to adapt an import routine to the geometric correction software.
4.2 Interior Orientation
The interior orientation consists of values and (if possible) residuals of angular increments, number of pixels per line,
number of channels, field of view (FOV), instantaneous field of view (IFOV), scan rate and rotation direction of the
rotating mirror or prism concerning the flight direction (clockwise or anti-clockwise). These data should be constant
during one mission and supplied together with a mission protocol.
4.3 Exterior Orientation
The exterior orientation defines the position and attitude of the sensor at each instant of time. There are six exterior
orientation parameters: three coordinates that define the position and three angles that define the direction of the line-of-
sight rays. The knowledge about the exterior orientation parameters is essential to solve the geometric correction
problem because they define the geometric linkage between the raw image data and the reference system to which they
have to be transformed.
4.4 Maps
Maps contain important geometric reference information but it has to be guaranteed that the map scale is large enough
to satisfy the needs. Normally maps are used to derive coordinates for ground control points. In this case it must be
possible to identify these points in both, the image data and the map. Sometimes maps are used to derive a digital terrain
model from contour lines. Maps provide also shape information (e.g. linear features) that can be used as constraints in
the geometric correction algorithm. An important information is the georeference (scale, reference frame, datum) of the
map. It is possible that the reference frame of the map does not correspond to the reference system needed for the
geometric correction of the hyperspectral data. In this case an appropriate transformation has to be applied to the
coordinates before they can be used.
4.5 Reference Orthoimage
Today orthoimages are available in many areas of our globe. Their potential is quite similar to those of maps. In
combination with a digital terrain model they serve for the derivation of ground control points. Like maps the scale (res.
the spatial resolution) must be large enough to satisfy the needs of precision. In comparison with the maps the
orthoimages contain image information instead of vector information. This makes it possible to apply matching
algorithms for automated control point derivation. For accuracy estimation it is important to know how the orthoimages
were compiled. Normally it is necessary that the rectification be done differentially whereas rectification based on
projective transformation can mostly not supply acceptable results (especially in rugged terrain).
4.6 Control Points
A control point consist of an identifier, two sets of coordinates and (if possible) residuals. Related to the image space a
control point has two plane coordinates in x' and y'. Related to the spatial reference system the same point is defined
with three coordinates X, Y and Z. It is necessary to calculate geometric corrections in this reference system X, Y, Z. It
is sometimes useful for point identification to have control point sketches which show the location of the points. As
mentioned above control points can be derived from maps and orthoimages but also from field surveys. The latter is of
course the most expensive way to get coordinates. But it is sometimes the only possibility to achieve high accuracy.
This is especially the case in remote areas where if at all only small scale maps or orthoimages are available.
4.7 Digital Elevation Model
A digital elevation model (DEM) describes the shape of the terrain in its three dimensions. The two main data structures
of a DEM are regular grids and triangulated irregular networks (TINs) (Hutchinson and Gallant, 1999). TINs are
96 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
