IEEE Geoscience and Remote Sensing Magazine - March 2021 - 66

would measure zero distortions if such data were available. Two score indexes for FR assessment are reported in
the following.

EXPERIMENTAL RESULTS
This section is devoted to the presentation of the experimental results.

QUALITY WITH NO REFERENCE
The QNR protocol takes separate measurements, each normalized in [0, 1] of spectral and spatial consistencies [115].
The motivation underlying this protocol is that the mutual
similarities between any couples of MS bands and each MS
band and PAN should be unchanged on average, varying
with the resolution, that is, before and after fusion.
The spectral distortion, D m , is calculated between the
LRMS images and the fused MS images [115]. Two sets of
interband UIQI values are calculated separately at low and
high resolutions. The differences of corresponding UIQI
values at the two scales yield the spectral distortion introduced by the pansharpening process.
The spatial distortion, D S , combines the UIQI values computed between each MS band and the PAN image degraded to the resolution of MS and again between
fused MS and FR PAN. The absolute difference, averaged
over all bands, between the corresponding UIQI values
yields the spatial distortion D S [115]. The spatial alignment between the original MS data and the PAN image
is of crucial importance to have an accurate evaluation
of this distortion index. The latter issue is considered to
be solved by the upsampling of the MS image to the PAN
scale. Thus, the implementation of this index can be performed considering the problem at the PAN scale, i.e., exploiting the upsampled MS image instead of the original
LRMS image.
The two distortions are then combined to provide a unique
quality index, referred to as QNR ! 60, 1@, with 1 being the
best attainable value:

DATA SETS
Three different scenarios acquired by three different sensors
are considered in this article. For all of the cases, both the
assessment at RR and at FR, following the indications drawn
in the " Quality Assessment of Fusion Products " section, are
provided, yielding six different test cases. The characteristics
of the employed data sets are detailed in the following.

QNR = (1 - D m) a (1 - D S) b, (25)

where usually a = b = 1.
HYBRID QUALITY WITH NO REFERENCE
The HQNR [117] borrows the spatial distortion index D S
from QNR and the spectral distortion index from Khan's
(K)
protocol [116], i.e., D m = 1 - Q2 n (\
MS .
MS ., R
MS), where \
is the MTF-filtered pansharpened MS image considering a
resolution ratio equal to R and R
MS is the original MS image
interpolated to the PAN scale, R times lower than the MS
scale. Again, this implementation is performed by exploiting
the upsampled MS image, i.e., working at the PAN scale.
(K)
It should be noted that D m strictly follows the consistency property and avoids possible drawbacks of D m , which
computes similarity indexes at different resolutions.
The two distortions are finally combined to yield the
unique quality index

(K)
HQNR = _ 1 - D m i ^1 - D Shb, (26)

where usually a = b = 1.
66

TOULOUSE DATA SET
This data set, exhibited in Figure 6(a) and (b), represents an
urban area of the city of Toulouse (France). It is acquired by
the IK sensor, which works in the visible and near-infrared
spectrum range. The MS sensor is characterized by four bands
(blue, green, red, and near infrared), and also a PAN channel
is available. The resolution cell is 4 m # 4 m for the MS bands
and 1 m # 1 m for the PAN channel. The resolution ratio R is,
therefore, equal to four. The radiometric resolution is 11 bits.
The size of an MS spectral band is 512 × 512 pixels. Thus, the
PAN image is 2,048 × 2,048 pixels. For the test case at RR, both
MS and PAN images are properly degraded, in agreement with
Wald's protocol. The ground-truth (GT) image is represented
by the original MS image, and the low-pass version of the MS
image has a size equal to 128 × 128 pixels.
COLLAZZONE DATA SET
This data set depicted in Figure 6(c) and (d) refers to a rural
area of Collazzone, a small town in Central Italy. It is acquired by the GE-1 sensor, which works in the visible and
near-infrared spectrum range. The MS sensor is characterized by four bands (blue, green, red, and near infrared),
and also a PAN channel is available. The spatial sampling
interval (SSI) is 2 m for MS and 0.5 m for PAN, respectively.
Thus, R is equal to four. The radiometric resolution is
11 bits. The size of an MS spectral band is 512 × 512 pixels. Therefore, the PAN image is 2,048 × 2,048 pixels. For
the test case at RR, both MS and PAN images are properly
degraded, in agreement with Wald's protocol as for the Toulouse data set.
NEW YORK DATA SET
The New York (NY) data set, provided in Figure 6(e) and (f),
represents an urban area of the city of New York with a more
relevant presence of high buildings with respect to European
urban scenarios, such as the one captured in the Toulouse
data set. These data are acquired by the (WV-3) sensor, which
works in the visible and near-infrared spectrum range. The
MS sensor is characterized by eight spectral bands (coastal,
blue, green, yellow, red, red edge, NIR1, and NIR2), and also
a PAN channel is available. The SSI is 1.2 m for MS and 0.3 m
for PAN, respectively. Therefore, R is equal to four. The radiometric resolution is 11 bits. The size of an MS spectral band
is 512 × 512 pixels. Thus, the PAN image is 2,048 × 2,048 pixels.
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

MARCH 2021

IEEE Geoscience and Remote Sensing Magazine - March 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2021