tanbul 2004
al Dimensions
2.950696
2.881100
2 915513
2.915515
2.948176
2.933445
2.935812
2.949563
t: IKONOS
ect power
ige fused by
ge fused by
rete wavelet
let transform
COMPARATIVE INFORMATION EXTRACTION FROM SAR AND OPTICAL
IMAGERY
Peter Lohmann, Karsten Jacobsen, Kian. Pakzad, Andreas Koch
Institute of Photogrammetry and GeoInformation (IPT)
University of Hannover, Germany
mailto:lohmann(gipi.uni-hannover.de
Commission 111, WGIII/6
KEY WORDS: Land cover, Image Interpretation, Radar, SAR,
ABSTRACT:
Presently EuroSDR, controlled by the Technical University of Berlin, is conducting a test on competitive information extraction
from state of the art airborne multi-polarised SAR imagery (C, X and L — band) and high resolution optical imagery of the same area.
The test envisages 3 stages, namely visual interpretation and map compilation, automatic object extraction and sensor fusion. Some
first results are shown. :
The interpretation results of three interpreters will be shown, two of them being skilled photogrammetric operators, the third having
only limited experience with SAR images. All interpreters have been given a principal theoretical course on SAR specific imaging
features and properties and training possibilities on a separate SAR scene together with map information which was not part of the
test.
A common interpretation and - for reasons of comparison - mapping key has been set up and on screen interpretation started using
the SAR scenes only, before the optical imagery was interpreted. Object extraction was conducted for linear objects like roads, rail-
ways or rivers and area objects like agricultural, forest or residential areas. In some limited areas also single buildings were extracted
in order to show the full potential of the imagery. Both interpretations (SAR and optical) have been compared according to accuracy
and completeness using the optical data as representative master because of missing reference data. Difficulties in identifying objects
by the interpreters will be discussed and compared It can be shown, that in most cases the completeness and correctness of linear and
area features as compared to the interpretation of optical data is satisfactory and that discrepancies between both interpretations may
be explained.
1l. INTRODUCTION optical images are used as reference. However each of the 3
interpreters used his own interpretation of the optical data as
Within an EuroSDR contest, organised by the Technical Uni- his own reference.
versity of Berlin, a competitive information extraction is per-
formed on both, state of the art airborne multi-polarized SAR Region Landscape Content
imagery (sce Table 1) and high resolution optical imagery of Trudering Fairgrounds
different tercain types: Oberpfaffenhofen Agriculture & Industry
Region A | Polarisation | Resolution Copenhagen Residential & Industry
Trudering X | none : om Fjordhundra Agriculture & Fons
Oberpfaffenhofen | L lexicographic | 3.0m Table 2: Test sites and their characteristics
Copenhagen C | Pauli 40m The 4 test sites (see Table 2) show a different terrain struc-
Fjordhundra C | Pauli 4.0m ture in terms of landscape contents, ranging from agricultural
Table 1: SAR images used for the different test areas to rural and industrial.
Two of the interpreters do have a long year experience in in-
The used data sets are described in detail by Hellwich et al. terpretation of aerial photographs but not with SAR images.
2002. The true-colour optical data available for each arca
was resampled to the pixel size of the corresponding SAR
images. The optical images have been acquired at different
dates than the SAR data, however the exact acquisition times
were not provided.
The test is foreseen for three stages namely:
- visual interpretation
- A automatic object extraction and
- extraction from sensor fused data
This paper will show some first results of the first stage, in
which visual interpretation by on-screen digitizing had been
performed. The interpretation achieved with the SAR images
will be compared with the results based on optical images.
Topographic maps have not been available, by this reason the
2. INTERPRETATION PROCEDURE
As has been previously shown by Albertz (1970) and Schnei-
der (1974) the quality of image interpretation is very much
affected by the experience of the interpreters (Fig. 1). This is
especially true for SAR image interpretation because of to-
tally different imaging characteristics.
Because of missing experience with interpretation of SAR
images the interpreters were trained for the specialities of
SAR by using a tutorial of CCRS and multifrequency SAR
images together with topographic maps of an area not being
part of this contest.
535