Environmental Engineering Reference
In-Depth Information
10.1
Introduction
an alternative to the IHS technique and works for an arbitrary
number of MS bands. PCA is analogous to the IHS scheme since
the Pan image is substituted to the first principal component
(PC1). Histogram matching of Pan to PC1 is mandatory before
substitution because the mean and variance of PC1 are generally
far greater than those of Pan. It is well established that PCA
performances are better than those of IHS (Chavez, Sides and
Anderson, 1991) since PCA is data dependent and thus capable
to fit data statistics. Consequently, the spectral distortion in
the fused bands is usually less noticeable than IHS, even if it
cannot completely be avoided. Generally speaking, if the spectral
responses of the MS bands are not perfectly overlapped with the
bandwidth of Pan, as it happens with the most advanced very
high resolution imagers, IHS- and PCA-based methods may yield
poor results in terms of spectral fidelity (Zhang, 2004). Another
CS technique reported in the literature is Gram-Schmidt (GS)
spectral sharpening, which was invented by Laben and Brower in
1998 and patented by Eastman Kodak (Laben and Brower, 2000).
The GS method has been implemented in the Environment for
Visualizing Images (ENVI) software, is widely used and produces
good
fused images. Its efficacy is mainly due the injection gain
that is proportional, for each band, to the covariance of the
synthesized intensity and the expanded MS band as reported
in (Aiazzi, Baronti and Selva, 2007). As a matter of fact, since
the sharp
P
and the smooth
I
have generally a different local
radiometry, spectral distortions can arise in the fusion results. A
mitigation of the consequent color changes can be obtained if
I
matches as much as possible the spectral response of Pan. This
result is achieved by designing
I
as a linear combination of theMS
bands by means of a set of coefficients according to the overlaps
existing among the spectral responses of MS and Pan images
(Tu
et al
., 2004). Such coefficients can be further optimized by
minimizing the distance between
P
and
I
,forexampleinthe
minimum mean square error (MMSE) sense (Aiazzi, Baronti
and Selva, 2007), utilizing a genetic algorithm (Garzelli and
Nencini, 2006) that optimizes the Q4 score parameter defined in
(Alparone
et al
., 2004b) or imposing MMSE constraints on the
multispectral images (Garzelli, Nencini and Capobianco, 2008).
The spatial resolution increase of recent satellite imagers is
impressive. IKONOS, QuickBird, Orbview, GeoEye-1, World-
View, KOMPSAT-2 are recording panchromatic (Pan) Earth
images with a spatial resolution that ranges from 0.41 m to 1 m.
Such other satellite imagers as Pleiades and GeoEye-2 are to be
launched in next future with a spatial resolution of 0.5 m and
0.25 m, respectively. All these sensors are therefore particularly
interesting for the analysis of urban areas exhibiting also small
spatial details. Most of these imagers also collect multispectral
(MS) data at a resolution that is four times lower than the Pan
image. In order to increase the resolution of MS to the resolution
of Pan data, remote-sensing image fusion techniques aims at
integrating the information conveyed by data acquired with dif-
ferent spatial and spectral resolutions. The most straightforward
goal is photoanalysis, but also automated tasks such as features
extraction and segmentation/classification in high spatial detail
areas have been found to benefit from fusion (Wahlen 2002;
Zhang and Wang 2004; Bruzzone
et al
. 2006; Colditz
et al
2006).
An extensive number of image fusion methods have been
proposed in the literature, starting from the second half of
the 1980s (Chavez, 1986). Most of them are based on a general
protocol inwhich high-frequency spatial information is extracted
from the Pan image and injected into the resampled MS bands
by exploiting different models. In general, the image fusion
methods described by this protocol can be divided into two main
families: the techniques based on a linear spectral transformation
followed by a component substitution (CS) and the algorithms
that exploit a spatial frequency decomposition usually performed
by means of multiresolution analysis (MRA). Under general and
likely assumptions, CS methods are insensitive to aliasing and
little sensitive to mis-registrations of moderate extent, unlike
MRA methods are. Conversely, MRA methods are less sensitive
to temporal shifts, i.e. MS and P acquired at different times
that may introduce spectral distortions, or color changes (Ehlers,
Klonus and Astrand, 2008).
10.1.1
Component substitution
fusionmethods
10.1.2
Multiresolution analysis
fusionmethods
Basically, CS techniques linearly transform the MS data set into
a more uncorrelated vector space. Then, one of the transformed
bands, usually the smooth intensity
I
,is
replaced
by the Pan
image
P
, histogram-matched to the
I
component itself, before the
inverse transformation is applied. This procedure is equivalent
to
inject
, i.e., add, the difference between
P
and
I
into the
resampled MS data set as shown in (Tu
et al
., 2001). One of
the first CS techniques that was applied is the intensity-hue-
saturation (IHS) method (Carper, Lillesand and Kiefer, 1990)
thatisappropriatewhenexactlythreeMSbandsaretobefused
since the IHS transform is defined for three components only.
When more than three bands are available, a generalized IHS
(GIHS) transform can be defined by including the response of
the near-infrared band into the intensity component
I
(Tu
et al
.,
2004). In this case,
I
is obtained by weighting the MS bands with
a set of coefficients whose choice can be related to the spectral
responses of the Pan and MS bands (Tu
et al
., 2004; Gonzales-
Audıcana
et al
., 2006). Principal component analysis (PCA) is
Although the spectral quality of CS fusion results may be suffi-
cient for most applications and users, methods based on injecting
zero-mean high-pass spatial details, extracted from the Pan image
without resorting to any transformation, have been extensively
studied to overcome the inconvenience of spectral distortion.
In fact, since the pioneering high-pass filtering (HPF) technique
(Chavez, Sides and Anderson, 1991), fusion methods based on
injecting high-frequency components into resampled versions of
theMS data have demonstrated a superior spectral fidelity (Wald,
Ranchin and Mangolini, 1997; Alparone
et al
., 2007). HPF basi-
cally consists of an addition of spatial details, taken from a
high-resolution Pan observation, into a bicubically resampled
version of the low resolutionMS image. Such details are obtained
by taking the difference between the Pan image and its low-pass
version achieved through a simple local pixel averaging, i.e., a box
filtering. Later improvements have been obtained with the intro-
duction of multiresolution analysis (MRA), by employing several
decomposition schemes, specially based on the discrete wavelet
Search WWH ::
Custom Search