2. EDUCATION IN GEOMETRIC IMAGE
PROCESSING
2.1 Teaching on reasons for geometric image
correction
The tradition in Hannover emphasises the role of
geometry for the practical application of remote
sensing techniques.
The didactic concept is permanently adapted to
practical requirements. It starts with pointing to the
fact, the main reason for geometric image processing
is, to derive geometric correct positions of the pixels,
which contain surface related grey value information,
in order to achieve a reliable geocoding of geometric
distorted remote sensing image data for the
correlation with GIS- or map-data.
Thus the full success of a remote sensing mapping
campaign depends on the ability of a proper
rectification of geometric distortions, in particular, to
avoid misinterpretations due to mismatching within a
GIS.
Main objectives in this context are the providings of
suited algorithms and software, necessary to produce
(digital) geometric corrected raster data of remote
sensing imagery in view of
- GIS Integration
- mosaicing,
- sensor comparison,
- updating etc..
This task has been solved by the development and
improvement of suited algorithms, which at least
allow to calculate 3 dimensional and not only 2
dimensional ground control point coordinate
information.
2.2 Leading to geometric approaches
The transformation of imagedata into a map or GIS-
coordinate system is an interpolation problem .
The interpolation process can be based on
- ground control point- and/or texture information,
- housekeeping data (like airborne GPS etc.) and on
- combinations of both types of data.
Ground control point data is derived from known
numeric coordinate values, from GPS, from maps
and/or from imagecoordinates of corresponding
points, which can be verified manually, but
increasingly by interactive digital (relative and
absolute) correlation techniques.
288
There are in principal 3 categories of algorithms to
solve the geometric problems for remote sensing
imagery:
- non-parametric interpolation methods
- (physical) parametric methods and
- combined approaches.
Non-parametric interpolation methods are namely
- polynomial equations,
- Spline functions,
- interpolation in a stochastic field, like moving
average, weighted mean, linear prediction etc..
For images of flat terrain 2 dimensional heuristic
polynomial equations for limited areas and stable
flight conditions for some cases already allow to
obtain sufficient accuracies for geometric
calculations.
The advantages of 2 dimensional polynomial
equations beside others are
- the didactic value for introducing into digital
geometric image processing,
- suitability for quick programming
- satisfying for limited areas of flat terrain
- support for approximate value determination
- support for blunder detection
Some disadvantages of 2 dimensional polynomials
are
- arbitrariness
- limited area of validation
- a blockadjustment based on arbitrary polynomial
equations of higher than first order shows an extreme
bad error propagation.
- restriction for 2 dimensions.
2.3 Education in improvement of algorithms
GIS systems still lack suited algorithms for geometric
as well as for radiometric manipulation of the data.
The development of suited algorithms for GIS
systems is still a real market gap. Therefore it is one
intention of the courses of study of remote sensing at
the University of Hannover, to prepare the students
for algorithm development and improvement.
In this context the formulation of strict geometric
algorithms for remote sensing imagery is a typical
sample for algorithm development. From the basic
idea a physical parametric solution is envisaged,
which allows to calculate
- the global and local behaviour of the sensor
position and attitudes,
- 3dimensional ground control point coordinates and
- computation of imagecoordinates for 3 dimensional
output raster data (resampling).