IEEE Geoscience and Remote Sensing Magazine - September 2015 - 28

MIguel A. VegAnzones
Gipsa-lab, Grenoble, France
(e-mail: miguelangel.veganzones
@gipsa-lab.grenoble-inp.fr)
geMIne VIVone
North Atlantic Treaty Organization (NATO) Science
and Technology Organization (STO) Centre
for Maritime Research and Experimentation (CMRE)
(e-mail: gemine.vivone@cmre.nato.int)
Abstract-Pansharpening aims at fusing a panchromatic
image with a multispectral one, to generate an image with
the high spatial resolution of the former and the high spectral
resolution of the latter. In the last decade, many algorithms
have been presented in the
literatures for pansharpening using multispectral data.
With the increasing availabilFUsINg Images wITh
ity of hyperspectral systems,
COmplemeNTaRy
these methods are now bepROpeRTIes,
ing adapted to hyperspecpaNshaRpeNINg helps
tral images. In this work, we
syNTheTIzINg a
compare new pansharpening
hypeRspeCTRal Image
techniques designed for hywITh a hIgh spaTIal
perspectral data with some
ResOlUTION.
of the state-of-the-art methods for multispectral pansharpening, which have been
adapted for hyperspectral
data. Eleven methods from different classes (component
substitution, multiresolution analysis, hybrid, Bayesian and
matrix factorization) are analyzed. These methods are applied to three datasets and their effectiveness and robustness are evaluated with widely used performance indicators.
In addition, all the pansharpening techniques considered in
this paper have been implemented in a MATLAB toolbox
that is made available to the community.
I. INTRODUCTION
n the design of optical remote sensing systems, owing
to the limited amount of incident energy, there are critical tradeoffs between the spatial resolution, the spectral
resolution, and signal-to-noise ratio (SNR). For this reason, optical systems can provide data with a high spatial
resolution but with a small number of spectral bands (for
example, panchromatic data with decimetric spatial resolution or multispectral data with three to four bands and
metric spatial resolution, like PLEIADES [1]) or with a high
spectral resolution but with reduced spatial resolution (for
example, hyperspectral data, subsequently referred to as
HS data, with more than one hundred of bands and decametric spatial resolution like HYPXIM [2]). To enhance the
spatial resolution of multispectral data, several methods
have been proposed in the literature under the name of
pansharpening, which is a form of superresolution. Fun-

I

28

QI WeI
University of Toulouse, IRIT/INP-ENSEEIHT
(e-mail: qi.wei@n7.fr)
nAoto YokoYA
University of Tokyo
(e-mail:yokoya@sal.rcast.u-tokyo.ac.jp)

damentally, these methods solve an inverse problem which
consists of obtaining an enhanced image with both high
spatial and high spectral resolutions from a panchromatic
image and a multispectral image. The huge interest of the
community on this topic is evidenced by the existence of
sessions dedicated to this topic in the most important remote sensing and earth observation conferences as well as
by the launch of public contests, of which the one sponsored by the data fusion committee of the IEEE Geoscience
and Remote Sensing society [3] is an example.
A taxonomy of pansharpening methods can be found
in the literature [4], [5], [6]. They can be broadly divided
into four classes: component substitution (CS), multiresolution analysis (MRA), Bayesian, and variational. The
CS approach relies on the substitution of a component
(obtained, e.g., by a spectral transformation of the data)
of the multispectral (subsequently denoted as MS) image
by the panchromatic (subsequently denoted as PAN) image. The CS class contains algorithms such as intensityhue-saturation (IHS) [7], [8], [9], principal component
analysis (PCA) [10], [11], [12] and Gram-Schmidt (GS)
spectral sharpening [13]. The MRA approach is based
on the injection of spatial details, which are obtained
through a multiscale decomposition of the PAN image
into the MS data. The spatial details can be extracted according to several modalities of MRA: decimated wavelet
transform (DWT) [14], undecimated wavelet transform
(UDWT) [15], "à-trous" wavelet transform (ATWT) [16],
Laplacian pyramid [17], nonseparable transforms, either
based on wavelets (e.g., curvelets [19]) or not (e.g., contourlets [18]). Hybrid methods have been also proposed,
which use both component substitution and multiscale
decomposition, such as guided filter PCA (GFPCA), described in Section II-C. The Bayesian approach relies on
the use of posterior distribution of the full resolution target image given the observed MS and PAN images. This
posterior, which is the Bayesian inference engine, has two
factors: a) the likelihood function, which is the probability density of the observed MS and PAN images given the
target image, and b) the prior probability density of the
target image, which promotes target images with desired
properties, such as being segmentally smooth. The selection of a suitable prior allows us to cope with the usual illposedness of the pansharpening inverse problems. The
variational class is interpretable as particular case of the
ieee Geoscience and remote sensing magazine

september 2015
http://gipsa-lab.grenoble-inp.fr
Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2015
