International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
space, grey value profiles over the edge should be measured in
the image. The response to the edge in the image will not be so
sharp like on the ground. The inclination of the grey value
profile in this location includes the information about the
effective pixel size.
Figure 1: edge analysis
upper left: IKONOS pan
1m ground pixel size
upper right: SPOT 5
5m ground pixel size
lower left: IRS-1C
5.8m ground pixel size
The same edge available in different space images (in figure 1
marked by red line) has been investigated for the edge response.
gray C—O
value
X
—— —
Figure 2: edge analysis
left: grey value profile in object space
centre: grey value profile in image space
right: differentiation of grey value profile in image
point spread function
The differentiation of the grey value profile in the image leads
to the point spread function. The width of the point spread
function at 50% height can be used as effective pixel size. In
the area of Zonguldak, Turkey, different space images have
been analysed at the same location. Of course not only a single
profile has been used for the analysis but all possible profiles at
the edge.
nominal pixelsize | effective pixel size
ASTER 15m 16.5 m
TK 350 (10 m) 13m
IRS-1C 5.8m 6.9 m
SPOTS 5m 5m
KVR 1000 ( 1.4 m) mn
IKONOS pan I m 1.0 m
Table 1: effective pixel size determined by edge analysis
Only the digital images ASTER and IRS-IC do show an
effective pixel size larger than the nominal pixel size. The
TK350 and the KVR 1000 are originally analogue space photos.
The KVR 1000 was delivered digitized with 1.4m pixel size on
the ground and the TK 350 has been scanned with a pixel size
of approximately 10m. For analogue images of course it is the
question if the pixel size used for scanning corresponds to the
image resolution and so it is not astonishing if we do have here
larger differences between the nominal and the effective pixel
size. Sojuzkarta talks about a smaller pixel size for the Russian
space photos, but they are always a little optimistic.
3. IMAGE OVERVIEW
A simple comparison of the different space images available
from the area of Zonguldak gives a good impression about the
information contents. The Synthetic Aperture Radar (SAR)
image from JERS with 18m pixel size includes only some
rough information about the area (figure 3). The information
contents of Radar images cannot be compared with optical
images having the same pixel size. Also based on other data
there is approximately the relation of 5 between them — 18m
pixel size of a SAR images includes approximately the
information of an optical image with 90m pixel size. But this is
only a rough figure because some details can be identified very
well in SAR images. For example in the JERS-image in figure
3, the white spot in the upper left corner is a ship which can be
seen more clear in Radar images than in optical images. The
Landsat TM image (figure 4) includes with the 30m pixel size
quite more topographic details like the JERS image.
FEA WP
Figure 3: Synthetic
Aperture Radar JERS
Area Zonguldak,
Turkey
Ground pixel size
18m
A comparison between the colour image of Landsat 7 TM
(bands 432) and the panchromatic Landsat image with 15m
pixel size shows more or less no advantage of the higher
resolution of the panchromatic band, but in general, the quality
of the panchromatic Landsat image is not so good in relation to
other space images with 15m pixel size like for example
ASTER. ASTER images do have usually a very good contrast.
The combination of the green, red and near infrared band
includes the advantage of a very good differentiation also in the
forest areas. The colour images in the visible spectral range
(red, green, blue) are influenced by the low contrast of the blue
channel caused by the stronger atmospheric scattering of the
shorter wavelength. In addition the blue and green band do have
a stronger correlation, that means in addition to the green and
red band the blue band includes quite less information like the
near infrared. By this reason also the visible and near infrared
(VNIR) combination of Landsat is shown in figure 3.
Landsat TM images are optimal for the classification of the land
use. The large pixel size of 30m is averaging the details which
are causing problems for an automatic classification based on
images with a small pixel size. But only few details required for
the generation of a topographic map can be identified.
Highways, especially in forest and agricultural areas can be
seen, but no more details. Under the condition of a geometric
mapping accuracy not better than 1 pixel and a requirement of
584
Interi
