International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
achieved at the cost of loosing some original advantages of 
satellite remote sensing. 
1.2. Combined Sensor Resolution and Data Fusion 
The engineer’s answer to the above-described dilemma is a 
compromise. In current and planned systems, lower resolution 
multispectral channels are often complemented by a single 
higher resolution panchromatic band, which provides a 
“representation” of the desired resolution for the multispectral 
channels. With this combined resolution concept, the acquired 
data volume is kept within acceptable bounds, while additional 
spatial details are still made available (Table 1). 
Several sensors with such constellations are already in orbit 
(SPOT, IRS-ID, Landsat 7), while others are planned (e.g. for 
1999 Ikonos-2, Quickbird-1, OrbView-3). Currently, the Indian 
remote sensing satellite IRS-ID offers 5.8m spatial resolution in 
the panchromatic range. The planned new satellite generation 
will provide lm panchromatic and 4m multispectral resolution 
(because of the data quantity problem, this extremely high 
resolution allows only restricted swath widths between 8 and 
22km only). 
Instead of using the panchromatic information separately, there 
is also the option to fuse the high resolution panchromatic band 
with the low resolution multispectral data to improve the spatial 
resolution of the latter (Figure 1). The aim of the fusion 
procedure is to produce high quality data which contains the 
characteristic of both the multispectral information (object 
identification) and the spatial detail (object localisation and 
texture). The “permission” to merge the data is seen in the fact 
that edge localisation (as detail manifestation) is more or less 
identical in different spectral bands and “only” varies in 
strength and polarity (Schowengerdt, 1980; Tom, 1986). 
2. CATEGORIZATION OF DATA FUSION 
TECHNIQUES 
During the last 20 years, many fusion algorithms were 
developed and documented. While the data quantity problem 
and the idea of data fusion stimulated the deployment of 
combined resolution sensors, the increasing availability of 
combined resolution data stimulated the further development of 
fusion algorithms. The intentions of algorithm developers or 
users are quite different. They encompass the desire for simple 
visual enhancements, as well as the demand to reconstruct the 
theoretically “real” high resolution multispectral image (the true 
image) as well as possible. 
Fusion techniques may be categorised into 3 groups, depending 
on how the panchromatic information is used during the fusion 
procedure. 
2.1. Fusion Procedures Using All Panchromatic Band 
Frequencies 
This category includes simple (partially old) band-arithmetic 
techniques, such as addition and multiplication, as well as the 
commonly used and well known component substitution 
techniques such as IHS transformation, principal component 
substitution (PCS) and the more rarely applied regression 
variable substitution (Haydn et al., 1982; Carper et al., 1990; 
Shettigara, 1992; Pellemans et al., 1993). Also the Brovey or 
color normalizing algorithm (Roller and Cox, 1980; Hallada 
and Cox, 1983), which holds an intermediate position between 
band-arithmetic and component substitution techniques, can be 
classified into this group. 
Because all procedures of this category make use of all 
frequencies of the panchromatic channel (which include also 
“spectral” information components related to the lower 
resolution multispectral images), they may produce spectral 
distortions in the result, which depend on the degree of global 
correlation between the panchromatic and the multispectral 
channels to be enhanced. Especially for NIR or SWIR channels, 
this global correlation is usually low, such that fusion 
procedures are bound to fail. Nevertheless, the IHS trans 
formation and the Brovey algorithm are often used to fuse the 
panchromatic band with multispectral channels within the 
visible spectrum (i.e. channels which are highly correlated with 
the panchromatic band). 
2.2. Fusion Procedures Using Selected Panchromatic 
Band Frequencies 
Fusion techniques classified in this group overcome the 
limitation described above, because only the additional high 
resolution spatial information (i.e., the high frequencies) of the 
Î 
multispectral dataset 
(low resolution) 
fusion procedure 
î 
panchromatic band 
(high resolution) 
î 
multispectral dataset 
(high resolution) 
Fig. 1. Concept of data fusion for resolution enhancement (modified after Pradines, 1986).