es: further
> realised.
ATION OF
)
0 make is
raster data
rived from,
ferences to
on of raster
nality.
g can utilise
IS, it is still
held in one
function to
. basis will
ctor object,
tability for
iS a single
genuinely
sensed data
processing
operations.
distinction
ed imagery
.E IMAGE
IS data can
processing
pint (GCP)
processing
fied. If the
> GIS data,
e GCPs can be extracted from a vector
database (typically road junctions)
e Vector data can be overlaid on an image to
verify the rectification process
e A DTM can be used to create an ortho-
image by differential rectification
4.2 Image classification.
This is, perhaps, the main task performed by
image processing system, and can be enhanced
in a number of ways by the use of GIS data:
e The use of vector objects for training sets.
Instead of identifying and digitising training
sets on the image, an existing vector
database (with attributes held in a relational
database) can be used to select objects to be
used for training the classifier.
e Object classification. If the image to be
classified is comprised primarily of
homogeneous objects (such as fields), it is
more efficient to classify the image on an
object-basis, with the objects defined by a
vector dataset (ie. a single classification is
given to each field, rather than to each pixel
within the field, eg. Pedley and Curran,
1991). This gives the classification greater
reliability (as untypical pixels - such as
mixed pixels at the field edges - can be
easily filtered out).
* Terrain illumination correction. Images with
illumination differences caused by terrain
can be corrected by applying an
illumination model generated from a DTM
(eg. Jones et al, 1988). (Incidently, if an
image is to be used for an image-map, it
may be necessary to use this technique to
‘change’ the illumination direction from
south-east to north-west, to avoid the
'pseudo-relief' effect).
* Knowledge-based classification. The latest
developments of classification algorithms
(eg. Janssen and Middelkoop, 1992) begin
to use GIS data for improving classification
by applying roles. Typical GIS datasets that
could improve the classification process
include a DTM, rainfall, slope/aspect and
existing land-use information.
387
4.3 Masking operations.
Vector objects can be used for a wide range of
image processing functions to mask areas for
processing. A simple example of this would be
to use the coastline to mask out the sea or land
when applying special processing algorithms to
the other. Linear vector geometries
representing boundaries of spatial changes in
spectral response can be used to define fuzzy
edges for mosaicing operations.
All of these functions have been investigated
and shown to be worthwhile enhancements to
current image analysis procedures. The main
reason why these have not been more widely
implemented, however, is the difficulty in
building a comprehensive range of vector
handling functionality into most image
processing packages.
5. FUNCTIONS WHICH ENHANCE
VECTOR OPERATIONS.
There are relatively few GIS functions that can
be enhanced directly by the use of remotely
sensed imagery. (However, data derived from
imagery - such as a land-use classification -
may be used routinely for a wide range of
applications). GIS functions using remotely
sensed imagery are:
5.1 Image-map backdrop.
Images are being increasingly used in areas of
the world where adequate base mapping does
not exist. This has the advantage of being
cheaper than vector mapping, more up-to-date
(and easily updated by acquiring the latest
imagery), and often showing ground features
that are not well represented by conventional
map symbology. These maps are often quite
adequate for locating "foreground" information,
such as pipes and cables.
5.2 Database update (eg. heads-up digitising).
This is the one area of image processing/GIS
integration that is often cited. Imagery -
particularly large scale aerial photography - is
widely used for updating a wide range of GIS
datasets from base maps to thematic overlays.
