752
October 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
Figure 6 shows a band-by-band comparison between
the original high resolution MS image, binned low res-
olution MS image, and pansharpened high resolution
MS image, taken under 18 lux and 100 ms. From Figure
6 we observe: (1) the SNR in each spectral band of the
pansharpened MS image is significantly higher than
that in its corresponding original high resolution MS
band; and (2) the gray value variations in each spec-
tral band of the original high resolution MS image, the
binned low resolution MS image, and the pansharpened
MS image are almost identical, if the noise effect and
the resolution difference are ignored. This also shows
that the pansharpening did not introduce any notice-
able spectral distortion.
For the images taken under the other three illumina-
tion and exposure conditions, similar qualitative eval-
uation and comparison results were observed. Due to
the page limitation of this article, these results cannot
be presented. However, the quantitative evaluation re-
sults of all four image sets are presented.
Q
uantitative
C
omparison
To measure the SNR, color quality, and spatial resolu-
tion of the images before and after pansharpening, the
industry standard software tools Matlab and Imatest
were used. The SNR function of Matlab is commonly
used by the imaging industry to calculate the SNR of an
image. Imatest is the standard software tool for image
quality measurements. Both are widely used by imag-
ing industry in research, development, and manufac-
ture of new sensors.
SNR
The SNR of an image is defined as the ratio of average
image signal to average image
noise. For calculating SNR, the
average pixel value of a smooth
gray area in the image is used
as the average image signal val-
ue, and the standard deviation of
the pixel values of the same gray
area is used as the average image
noise value (Scientific Volume Im-
aging, 2016).
In this study, the SNR is mea-
sured in four different gray ar-
eas of the Noise Contrast Chart
(15739) (Imatest, 2016a) (Figure
7). In order to reduce random er-
rors, the four SNR measurements
were averaged together to produce
one SNR for each image band. For
MS images, the SNRs in the four spectral bands were
each measured and calculated, and then averaged to-
gether to output one SNR for each MS image (Figure 8).
From Figure 8, it can be seen that the SNRs of all the
pansharpened MS images are about two times those of
the original high resolution MS images, and they are
about the same to or slightly higher than those of the
original high resolution Pan images.
Figure 7. Four gray areas in the Noise Contrast Chart (15739) used
for SNR measurements.
Figure 8. Average SNRs of the original high resolution MS (HR MS), original high resolution Pan (HR
Pan), and pansharpened high resolution MS (PS HR MS) image for each image set.
40 lux × 20 ms 250 lux × 20 ms 18 lux × 100 ms 18 lux × 250 ms
HR MS
14.6
26.1
13.6
20.5
HR Pan
23.6
33.2
27.8
41.3
PS HR MS
26.3
41.1
30.1
37.8
0
5
10
15
20
25
30
35
40
45
SNR
Image Set
SNR Comparison
“From Figure 8, it can be seen that the SNRs of
all the pansharpened MS images are about
two times those of the original high resolution
MS images, and they are about the same to
or slightly higher than those of the original high
resolution Pan images. ”
