IMAGE RESTORATION OF CALIBRATION AND VALIDATION FOR KOMPSAT-2
Dong-Han Leea, D.C. Seo, J.H. Song, J.H. Chung, S.Y. Park, M.J. Choi, H.S. Lim
a
Satellite Data Information Team, Korea Aerospace Research Institute (KARI), 45 Eoeun-dong Yuseong-gu, Daejeon
305-333, South Korea – dhlee@kari.re.kr
Commission I, WG I/1
KEY WORDS: Calibration, Passive Optical system, Image restoration, MTF, Radiometry, Camera, Sensor, KOMPSAT-2
ABSTRACT:
The image restoration of Calibration and Validation (Cal/Val) for the KOMPSAT-2 (KOrea Multi-Purpose SATellite-2) has six
parameters; Linearity, Video processor gain/offset, Non-uniformity correction, De-noising, Butting zone, SNR and additionally
MTF/MTFC. Before launched, Cal/Val site, equipments, code and procedure for the image restoration of Cal/Val and MTF/MTFC
have been defined and developed. After launched, KARI Cal/Val team has done the image restoration of Cal/Val and MTF/MTFC
for the KOMPSAT-2, and then can guarantees the KOMPSAT-2 radiometric image qualities and MTF for Users.
Linearity: < 4% (5%-95% sat.)
SNR: > 100
1. INTRODUCTION
1.1 Overview
The next list explains MTF specification of the KOMPSAT-2;
Line rate: up to 7100 lines/sec
- Default value (at Nadir imaging): 6800 (PAN), 1700
(MS)
MTF (@ Nyquist freq.): ~8%
Yaw steering accuracy shall < 0.05 degree (3σ)
After KOMPSAT-2 launched, the Cal/Val for KOMPSAT-2
image quality has been doing and implementing the
KOMPSAT-2 image data processing system with it in KARI.
Generally, because the present remote sensing satellite
technique cannot satisfy user’s requirements for image quality,
the Cal/Val for image quality must be carried out directly after
launch before distributing the imagery to the users. In the broad
concept of Cal/Val, the Cal/Val of the remote sensing satellite
can be divided into two parts if we recognize the technical gap
between the satellite technique and the user requirements;
Cal/Val to validate and verify the requirements of satellite, and
Cal/Val and image restoration to guarantee the image data
quality for the users. After launch, we can immediately
understand the gap between the real Cal/Val and our own way
from the new and different phenomenon and our mistakes form
analyzing the KOMPSAT-2 image data. The first gap between
them is the radiometric Cal/Val; Linearity, Video processor
gain/offset, Non-uniformity correction, De-noising, Butting
zone and SNR (Signal to Noise Ratio). The second gap of it is
MTF (Modulation Transfer Function) / MTFC (MTF
Correction). The last gap of it is the geometric Cal/Val; Geoaccuracy, Band-to-Band registration, Planimetric accuracy, etc.
(for more detailed; Seo 2007)
2. RADIOMETRIC CAL/VAL
2.1 De-noising
The noise value of the KOMPSAT-2 specification, 1% after
NUC, is too high to be accepted by the Users. Noise can be
divided by the random noise and the pattern noise. The random
noise can directly affect the SNR, and be reduced by some filter
algorithm. The KOMPSAT-2 IRPE (Image data processing
system) has a wavelet filter to reduce the random noise in the
KOMPSAT-2 image data. The pattern noise within the
KOMPSAT-2 noise requirement level has been found out after
KOMPSAT-2 launch newly; Diagonal line, Vertical line and
Horizontal line. Additionally, the pattern noise may depend
upon the input radiance (DN; Digital Number). The vertical and
horizontal noise may appear more at less than 200 DN
(Residual NUC). The diagonal line appears periodically, and
the vertical and horizontal noise non-periodically.
This paper explains de-noising, non-linearity and butting zone
of the radiometric Cal/Val and MTF/MTFC of KOMPSAT-2,
Lee 2007 does the overview of KOMPSAT-2 Cal/Val and the
basic Cal/Val results. Seo 2007 paper explains the Geometric
Cal/Val of KOMPSAT-2, and Song 2007 does NUC
Radiometric Cal/Val.
The diagonal line has been removed by the FFT algorithm
developed from the periodic pattern of it (Figure 1). In Figure 1,
the non-periodical horizontal line can be found out easily after
the diagonal line removed.
a.
1.2 KOMPSAT-2 radiometric and MTF specification
The next list explains the radiometric specification of the
KOMPSAT-2 MSC (Multi-Spectral Camera);
Non-uniformity correction before compressing
Residual Non-Uniformity: < 1%
57
Diagonal line noise
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing 2008
The Linearity value of the KOMPSAT-2 specification, 4% from
5% to 95% saturation level, is too high to be accepted by the
Users. The non-linearity of KOMPSAT-2 image data has been
analyzed with the uniform areas of several DN values, so that
three regions for grey level (DN) can be divided in Figure 3.
The vertical and horizontal noise can be removed with each
region of the next;
Region 2: HF NUC in MSC on-board
Region 1: Algorithm for removing Residual NUC (Vertical
line); Stronger Non-Linear CCD Response in Region 1
Region 3: Difference between Even and Odd pixel value
Figure 1. Up: Raw image data, Bottom: Diagonal line removed
The non-linearity can affect the next;
Butting zone
Vertical & Horizontal line
MS-band Fluctuation
Normalization with Video gain & NUC
MTF Curve
MTF Correction
Absolute radiometric Cal/Val
2.3 Butting zone
Figure 2. FFT filter in frequency domain to remove the vertical
and horizontal noise line
KOMPSAT-2 MSC PAN has three CCDs of 5200 pixels. There
are two butting zone between CCDs that decrease the brightness
gradually (Figure 5). Because there is some non-linearity in the
butting zone, we have developed three more algorithms to
reduce it; Scatter plot, BSM (Butting zone Smoothing Method)
and Dispersion method.
b. Non-periodical vertical and horizontal noise
To remove the non-periodical vertical and horizontal noise line,
the FFT method in frequency domain has been used. In Figure 2,
the FFT filter has been analyzed carefully only to remove the
vertical and horizontal noise line without any blurring.
100 (overlap)
PAN2
PAN4
50%
72
Figure 3. Left: Vertical & Horizontal Pattern noise, Bottom:
Removed by FFT method
72
Figure 5. The brightness degradation in the Butting zone
2.2 Non-Linearity
DN
Region 3
Region 1
Region 2
Figure 6. Up: The butting zone between PAN2 & PAN4,
Bottom: Corrected
Radiance
Figure 4. Non-linearity of KOMPSAT-2 image data
58
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing 2008
PAN2
PAN4
Get a Dispersion Value from this part
1.01
Get Gain/Offset of every pixel with
the Dispersion Value of it
1
0.99
0.98
0.97
0.96
0.95
0.94
Figure 7. Non-linearity in Butting zone
0.93
0.92
0.91
9500
The butting zone has the difference in grey level, contrast,
clearness, discontinuity, etc. between each CCDs, and strong
non-linear CCD response (Figure 7). To reduce the butting zone
issues, the relation, the scatter plot between PAN2 / PAN4 and
PAN3 / PAN5 has been searched for all grey level possibly
with so many KOMPSAT-2 image data (Figure 7 right). In
Figure 7, if KOMPSAT-2 MSC be perfect system, it should be
red line, but we got the blue line of the non-linearity depended
upon the grey level., The right image in figure 8 is corrected by
the result of the scatter plot between PAN2 and PAN4, but there
is still remnant in butting zone.
9600
9700
9800
9900
10000
10100
10200
10300
10400
10500
Figure 10. The principle of Dispersion method
In Figure 11, the left image is only corrected with LF NUC +
Scatter Plot, the middle image is corrected with LF NUC +
BSM + Scatter Plot, and the right image is corrected with LF
NUC + BSM + Scatter Plot + Dispersion method. The right
image can be clearer than the left and the middle. If LF NUC +
BSM + Scatter Plot + Dispersion method can be applied to the
KOMPSAT-2 image data, there are no more butting zone issues.
LF NUC+BSM+
ScatterPlot
LF NUC+ScatterPlot
LF NUC+BSM+
ScatterPlot+Dispersion
Figure 11. The result of BSM + Scatter Plot + Dispersion
method in butting zone
Figure 8. Up: Raw image data, Bottom: corrected by the scatter
plot
3. MTF AND MTFC
To remove the butting zone issues in all case of the
KOMPSAT-2 image data, two more algorithm has been
developed; BSM (Butting zone Smoothing Method) and
Dispersion method. To fully remove the butting issues, do the
next sequence; LF NUC → BSM → Scatter Plot → Dispersion.
In figure 9, the principle of BSM is the smoothing of the middle
DN value of all lines at the butting zone. The weak point of the
BSM is the case of the right of Figure 9.
3.1 MTF Value
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
1.0
2.0
1.0
0.9
0.8
0.8
PAN2
Relative Contrast
0.7
In case, Weak point
PAN4
After LF NUC
Mean value
0.6
0.6
0.5
0.4
0.4
0.3
6%
0.2
0.2
0.1
0.0
0.0
-0.1
-0.2
Figure 9. The principle of BSM
0.0
0.2
0.4
0.6
0.8
1.0
Lines/m
1.2
1.4
1.6
1.8
-0.2
2.0
Sep. 3, 2006
Figure 12. The beginning MTF value of the KOMPSAT-2 with
Siemens target at Stennis Space Center in USA
59
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing 2008
The MTF value of PAN at nyquist frequency is 8% in the
KOMPSAT-2 specification. The beginning MTF value was less
than 8% (~ 6%) (Figure 12), so that we has analyzed the factors
of the KOMPSAT-2 MTF; AOCS stability & bias, noise, TDI
line rate error and Yaw steering error.
F ul l Wi d th at H a lf Ma xim u m
1
0.9
N orm al iz ed Di fferenti ati on va l ue
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
FWHM = 2.5373 [Pixel]
0
-6
-4
-2
0
2
4
6
Pix el
Figure 15. Non-symmetry of the KOMPSAT-2 PSF
1.2
1.0
0.8
0.6
0.4
across
Figure 13. The analysis result of TDI line rate and Yaw steering
for MTF
0.0
-8
MTF (Along)
※ N ight Lam p im age in M ongol; Feb. 22, 2007
-2
0
2
4
6
8
MTF Along
0.8
Normalized value
2
-4
MTF Along at Nyquist = 6.8655%
The optimal PSF could be measured from nine Night lamp
images (Figure 17, Right), and then the MTF kernel (Figure 18)
has been generated by the Wiener filter (Figure 17, Left). The
size of the generated MTF kernel is 13 x 13.
0.7
Point Source
-6
Figure 16. PSF from the Night images
1
0.9
※ A fter upload A O C S K P D (Y aw steering)
along
0.2
0.6
0.5
0.4
0.3
4
0.2
6
Pixel
0.1
0
8
0
0.05
0.1
0.15
0.2
0.25
0.3
Normalized frequency
0.35
0.4
0.45
0.5
MTF (Across)
1
10
MTF Across
P
SF
PS
PSF
F
0.9
MTF Across at Nyquist = 7.0444%
12
h(x,y)
h
(x,y)
0.8
Inp
ut
Input
Input
Spatial=> frequ
fre quency
ency
6
8
Pixel
10
0.7
12
Normalized value
4
1.5
Transfo
Tran
sform
rm
Transfo
rm
0.6
OTF
OTF
O
TF
MTF
Convolution
Convolu
tion
Convolution
with
Output
utput
with O
Ou
tpu t
0.5
H(v1 , v2)
H(v1,
v2)
Q(v1,
Q
(v1, v2)
Q(v1,
v2)
hhII(x,y
(x,y)
y(x,y
(x,y)) ⊗ y(x,y)
y(x,y))
1
0.5
0.4
Inve
Inverse
MTF
Inverse
rse MTF
0.3
Inve
Inverse
rse
Inverse
Filter K
Ke
ernel
rnel
Ker
nel
H
HII(v1,
v2
(v1 , v2)
v2))
0
1
2
3
4
5
6
7
8
9
10
11
12
13
-0.5
hII(x,y
(x,y)
(x,y))
0.2
Inve
Inverse
Fourie
In verrse
se Fourier
Fourierr
0.1
0
2
a(x,y
a(x,y)
a(x,y))
Fourier
Fou
rier
rier
2
0
0.05
0.1
0.15
0.2
0.25
0.3
Normalized frequency
0.35
0.4
0.45
Transfo
Transforrm
m
Transform
freque
ncy=>spati al
freq uency=>spatial
0.5
-1
-1.5
Figure 17. Left: Wiener filter to generate the MTF kernel,
Right: MTF kernel (13x13) for MTFC
Figure 14. The MTF value from the Night lamp
In Figure 13, we have found out the effect of the TDI line rate
could be neglected, but the Yaw steering value was higher than
the KOMPSAT-2 specification (0.05 deg). After KARI AOCS
team uploaded the AOCS calibration data included with the
Yaw steering offset at Feb 14, 2007, the MTF value of
KOMPSAT-2 image data was measured again with the image
from the Night lamp (Figure 14), and then we got the around
‘7%’ of the MTF value.
3.2 MTF Compensation (Correction)
We have studied the difference between ‘7%’ from the Night
lamp and ‘8%’ from the KOMPSAT-2 specification, and then
the reason of it;
Non-symmetry of the KOMPSAT-2 MTF curve (Figure 15)
Non-linearity in less than 200 DN
Figure 18. Left: Raw image data, Right: MTFC image data
The KOMPSAT-2 IRPE has a framelet wavelet de-convolution
algorithm (Jeong) for the MTFC and the de-noising. In Figure
18, MTFC image data is better quality than raw one. The MTF
value of the MTFC image data is 11 ~ 12 %. But, there is a
minor problem of it; aliasing.
This difference can be compensated by MTFC, so that the
optimal Point Spread Function (PSF) for the KOMPSAT-2
image data has to be needed for it.
60
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing 2008
4. CONCLUSION
For one-half year after KOMPSAT-2 launch, KARI Cal/Val
team has carried out the Calibration and Validation to guarantee
the image quality of KOMPSAT-2 MSC image data, so that the
Users can be taken the KOMPSAT-2 image data with good
image quality now. Nevertheless, there is much room for more
enhancements of the image quality of the KOMPSAT-2 image
data; MTFC, Reduce noise, Geo-accuracy, Band-to-Band
registration, Absolute radiometric calibration, Users Manual,
etc.
REFERENCE
Lee, D., 2008, Summary of Calibration and Validation for
KOMPSAT-2, ISPRS2008
Seo, D., 2008, KOMPSAT-2 direct sensor modeling and
geometric calibration/validation, ISPRS2008
B. Jeong, M. Choi, H.O. Kim, A construction of symmetric
tight wavelet frames from quasi-interpolatory subdivision
masks and their applications, International Journal of Wavelets,
Multiresolution and Information Processing, Accepted.
61
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing 2008
62