International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
38
from a vector database according to their attributes. An
important aspect of registration is that all features do not need to
be extracted, only enough to carry out the registration with some
redundancy.
Point features are not suitable for matching from very different
types of data. A point which can be found by an interest
operator in an optical scene will not necessarily have a
conjugate in a SAR image (not least because of the speckle), nor
in a vector database. Linear features, especially intersections can
often be found in different data sets but in order to match them
an initial registration is usually necessary. These are therefore
used to refine the registration, if required, in our strategy. The
use of polygons allows automatic initial registration and the
possibility of refinement using points extracted from polygon
boundaries.
Reports on many of the issues discussed in this paper have been
published in papers which describe two systems developed by a
consortium led by University College London (UCL). A system
has been developed for automatic registration of optical images
as part of a Prototype Automatic Image Registration System
(PAIRS) developed by Earth Observation Sciences Ltd.,
University College London, University of Stuttgart and the
University of Oporto, for the Western European Union Satellite
Centre (WEU) (Dowman et al. 1997). The work in the PAIRS
project has been extended with all of the partners in the WEU
consortium plus KTH Stockholm and the Swedish Space
Corporation, to develop the automatic registration of images to
maps. This is the ARCHANGEL project funded under the
European Union Fourth Framework research programme
(Dowman, 1998).
The image to map process is illustrated with examples from the
ARCHANGEL project. SPOT panchromatic data is matched to
German ATKIS data. ATKIS is a digital database which is
derived from the 1:5000 topographic maps. The OPTISEG
algorithm was used to extract polygons from the SPOT data and
the methods described by Hild and Fritsch (1998) were used to
extract polygons from the ATKIS data. The matching was done
using the match polygon algorithm developed for
ARCHANGEL from Abbasi-Dezfouli and Freeman (1994) and
a. ATKIS data
the points extracted using the method on Newton et al. (1994).
Figure 2 shows the ATKIS data with selected features (roads,
forests and settlements) extracted and the original image with
the polygons extracted and the road network overlain on the
registered data.
Work on extraction of intersecting features for the
ARCHANGEL project can be found in Klang (1997 and 1998)
and a full account of the ARCHANGEL project in
ARCHANGEL (1999).
The remainder of this paper will concentrate on new
developments in polygon extraction and matching with
examples of matching SAR data with SPOT data.
3. POLYGON EXTRACTION
Polygons may be extracted from imagery using three basic
techniques: thresholding, identification of homogenous patches
and segmentation.
Thresholding is a simple technique but not considered to be very
reliable. Up until now, the threshold has had to be chosen
manually and contrast might change significantly over an image,
making the choice of a global threshold impossible. In addition,
there is always a lot of clutter after the thresholding. A method
has been developed in which an image is tiled to reduce global
differences and then to smooth the histogram within a tile to
develop a bimodal histogram and then to automatically identify
the threshold point.
Homogeneous patch extraction (Abbasi-Dezfouli and Freeman,
1994) is also straightforward but again is not consistent on all
types of images. The algorithm works by scanning the image for
homogeneous patches (i.e. regions with the same grey level) and
separating them from the background. Since there is always
some variation in patches, a tolerance value is set which
specifies a range in which the grey level values must fall.
b. SPOT image with extracted polygons (ATKIS motorway &
road polygons overlain)
Figure 2. Registration of a SPOT image with map data using ARCHANGEL software.