International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
à : Sensor- Sensor- GCP accuracy AS
A ui Scanning A : Numbers : GCP measurement
Image Date of acquisition Azimuth | Elevation (m)
mode of GCPs method
(deg) (deg)
Geneva_Q 2003-07-29 Reverse 286.4 77.6 67 0.3-0.5 Orthoimage / laser DTM
Geneva I West 2001-05-28 Forward 253.6 67.2 34 0.3-0.5 Orthoimage / laser DTM
Geneva I East 2001-05-28 Reverse 240.2 61.6 44 0.3-0.5 Orthoimage / laser DTM
Thun I 49 000 2003-12-11 Reverse 140.35 62.78 25 0.2-0.3 GPS
Thun I 49 100 2003-12-11 Reverse 66.41 63.56 25 0.2-0.3 GPS
Thun I 51. 000 2003-12-25 Reverse 180.39 62.95 24 0.2-0.3 GPS
Thun I. 51. 100 2003-12-25 Reverse 72.206 82.13 24 0.2-0.3 GPS
Thun I 54. 000 2003-12-25 Forward 128.17 82.62 24 0.2-0.3 GPS
Table 1. Specifications of used satellite images and respective GCPs (Q stands for QUICKBIRD and I for IKONOS).
2. IMAGE ANALYSIS
2.1 Radiometric Quality
HRS usually employ TDI technology. All IKONOS and QB
images have been acquired using 13 stages of the TDI. A
higher number of stages would increase the signal but also the
danger of saturation, especially for bright objects. TDI results
in smoothing and a reduction of the MTF. MTFC is always
applied by SI and although in the QB metadata nothing is
mentioned, it is fairly probable that a similar process is
applied. DRA is optional with IKONOS, but with QB although
again nothing is mentioned in the metadata, it seems that it is
applied by default (this is indicated by the respective
histograms which show saturation in the maximum grey value
of 2047). The histograms of both IKONOS and QB show that
only 8-9 bit are essentially used, while the blue channel has
the smallest range of grey values.
The noise characteristics of the images were analysed and
quantified using the standard deviation of the gray values in
homogeneous (Lake of Geneva, Lake of Thun) and
inhomogeneous areas (large image parts without homogeneous
areas). The use of homogeneous areas is justified as noise is
especially visible in such areas, whereas the use of
inhomogeneous areas allows an analysis of the noise variation
as a function of intensity and when homogeneous areas are
missing. Baltsavias et al. (2001), in their first assessment of
IKONOS Geo, give a short description of the method utilised
for noise estimation. The method has been modified regarding
noise estimation in inhomogeneous areas, in order to adapt
computation of the standard deviation according to the number
of significant samples in each bin (grey level range).
Homogeneous areas existed only in the IKONOS East image of
Geneva and the eastern Thun stereo pair. In QB, due to wind,
the water surface was not homogeneous and could not be used.
The mean standard deviation is computed out of the N96 (here
8576) smallest percentage of samples. According to Table 2,
the noise in the Thun images is slightly less than in Geneva
and the MS exhibit less noise than the PAN ones, possibly due
to the 4 times larger pixel size. Considering the fact that the
| I-bit data represent actually only 8-9 bit, the noise is quite
high for PAN, a fact that could be verified visually by strong
image contrast enhancement.
Estimation of noise in inhomogeneous areas uses as input a
range of standard deviations in each bin, based on which a
percentage is computed. The standard deviation in
homogeneous areas is used to compute the input range. For the
Geneva IKONOS images, the values of the input range were
set to 3.5 for PAN and 1.5 for MS. For the QB PAN, the range
has been empirically set to 1.7. Table 3 shows the results for
the PAN channels, whereby the values for IKONOS are
average values. Table 3 indicates that noise is intensity
dependent for all images, however for QB the noise increases
less with intensity. When the number of samples in a bin is
less than 50, no value is given. The lower noise of QB may be
due to a better preprocessing of the QB images, or due to the
imaging conditions (e.g. higher elevation), or due to the fact
that QB while scanning the scene, e.g. from North to South
continuously rotates from South to North in order to achieve
the nominal pixel size for PAN, thus oversampling. But it can
also be accidental, or due to uncertainties in noise estimation in
inhomogeneous areas. Thus, more tests with QB images
involving also homogeneous areas are needed. For the MS
channels, in both IKONOS and QB, the noise pattern is similar
to PAN, however due to the lower dynamic range (shorter
integration time), less bins have a significant number of
samples.
Ikonos Red | Green Blue NIR PAN
images
Geneva East 1.89 D 34 235 2.02 5.05
Stereo A | 1.54 1.98 2.2] 1:75 4.20
Thun
Stereo B] 161i 1.77 1.93 1.66 4.13
Thun
Table 2. Noise estimation for homogeneous areas in
: IKONOS images.
PAN 0 — |128— | 256 — | 384 — | 512 - | 640 — | 768 —
Scenes 127 | 285 | 383 511 639 | 767 | 895
Geneva ] - 346 | 3007 | 403 | 420 | 561 6.26
Geneva Of [26 | 1.35 [138 | 1.33 [147 | 214 | 203
Thun EST | K95 [3.26 (5.54 - - -
stereo
Thun 1,82 {1,38 | 2:53-| 2.99 | 3,47 -| 4/59 -
triplet
Table 3. Noise estimation for inhomogeneous areas and
different grey value ranges (bins) in PAN images.
All images were found to exhibit artifacts, which were visible,
especially in homogeneous area and/or after strong contrast
enhancement. Stripes in flight direction due to imperfect
calibration of the sensor elements. Strong reflections in both
PAN and MS images, which lead to saturation of the signal and
loss of information. Spilling (Fig. 1(a), IKONOS, l(c) QB) of
bright target response in neighbouring lines in the flight
direction, visible almost exclusively in the PAN images and
blooming (Fig. 1 (b), IKONOS).
1251
