Image Processing Reference
typically result in multispectral images having the highest spatial resolution available
within the data set. They are applied on data sets comprising multispectral images
B kl at a low spatial resolution l and images A h at a higher spatial resolution h but
with a different spectral content. Examples of such data sets are SPOT-XS (3 bands,
20 m) and SPOT-P (panchromatic, 10 m) images, SPOT-4 (3 bands at 20 m) and the
band XS2 (10 m), IRS-1C (3 LISS bands, 23.2 m) and a panchromatic band (5.8 m),
or IKONOS (4 multispectral bands at 4 m) and a panchromatic band (1 m).
In this section, we discussed some of the methods which claim to provide a
synthetic image close to what would be observed by a similar multispectral sensor
with a better spatial resolution (Ranchin and Wald 2000b ). Those methods which
only provide a better visual representation of the set of images (Carper et al. 1990 )
are not addressed here.
Wald ( 2002 ) groups the synthetic merging methods under three categories:
Projection and substitution methods: These project original data sets into another
space, substitute one vector by the higher resolution image, and then inverse
projection into the original space. Examples of these methods is IHS (Intensity,
Hue, and Saturation) method (Carper et al. 1990 ).
Relative spectral contribution methods: These include the Brovey transform
(Pohl and Van Genderen 1998 ) which can be applied to any set of images, and
the CNES P+XS method (Anonymous 1986 ) developed for SPOT imagery.
It should be noted that the Brovey transform does not represent well synthetic
merging methods because of its poor principles in construction (Wald 2002 ).
Nevertheless, it is often used.
Methods relevant to the ARSIS concept: In these methods a scale-by-scale
description of the information content of both images is generated to facilitate a
high-quality transformation of the multispectral content (Ranchin and Wald
2000b ; Ranchin et al. 2003 ). The HPF (High Pass Filtering) method is an
example of these methods but does not usually provide quality transformation of
the multispectral content.
Synthetic Merging Methods
Let's denote the acquired images of lowest spatial resolution by B l , and the images
of highest spatial resolution by A h . The subscripts l and h denote the spatial resolu-
tion of images B or A , i.e. lower and higher resolutions respectively.
The fusion methods aim at constructing synthetic images B* h , which are close to
what would be observed by a similar multispectral sensor with a better spatial reso-
lution. The methods perform a high-quality transformation of the multispectral
content of B l , when increasing the spatial resolution from l to h . The general problem
is relevant to the fusion of representations and is the creation of a new representation
B* from the original representations A and B (Wald 2002 ):
B f AB