International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004
resolution multispectral imagery. This means that a pan-
sharpened image combines the spatial and spectral
characteristics of both images. Many algorithms and techniques
have been developed to sharpen imagery successfully, among
which the Principle Components Transformation was selected.
4.2 The preprocessing of the images
The images to-be fused should fulfill some conditions, to be
projected to the same coordinate system and to cover exactly
the same area. Although in this case the images to-be fused
were orthorectified it is also necessary to co-register the
multispectral to the panchromatic orthoimage. Only with the
co-registration of the low-resolution image to the high-
resolution image an absolute correspondence of the pixels with
the same coordinates is accomplished.
Additionally, the images to-be fused should have a radiometric
correlation between them. For this reason before the fusion, a
matching of the histogram of the panchromatic orthoimage to
the histogram of the multispectral orthoimage is essential. This
process guaranties that the two images have the same contrast
and brightness.
4.3 Fusion with the Principal Components Transformation
The Principal Components Transformation (PCT) is a
commonly used algorithm for the fusion of imagery. It
calculates the principal components of the multispectral image
and transforms the set of bands into pseudo-bands with the
same total, but a different distribution of variance. The first
principal component represents the largest variance, is
considered to resemble spectrally the panchromatic band and
thus it can be replaced by it. From the new PC image through a
PC-inversion the pan-sharpened image is generated.
Figure 4. The pan-sharpened image
44 Evaluation of the spectral quality of the pan-sharpened
image
Evaluating the spectral quality of the result of the fusion is a
very important process. The pan-sharpened image should
preserve the spectral characteristics of the images that were
produced. This involves a spectral comparison between the pan-
sharpened image and the multispectral orthoimage.
In order to be able to compare these images they should have
the same resolution. The 1-m pan-sharpened image was
degraded to 4-m. Apart from this it was essential to implement
a histogram matching of the pan-sharpened image to the
multispectral orthoimage so as to be radiometrically similar.
To assess the spectral quality of the pan-sharpened image
certain criteria were used:
The bias, which is the difference between the means
of the original multispectral orthoimage and the pan-
sharpened image. Ideally, the bias should be null.
The correlation coefficient between the original image
and the pan-sharpened image. It shows the similarity
in small sizes structure between the two images. It
should be as close as possible to 1.
The standard deviation of the difference image, which
globally indicates the level of error at any pixel.
Ideally, it should be null (Wald et al., 1997).
The NDVI index was applied to both images, which
detects vegetated areas. The original and the pan-
sharpened image images should have the same
spectral reaction to the specific index meaning that the
NDVI images should spectrally similar. Thus the bias,
the standard deviation and the correlation coefficient
between the NDVI images were computed (Tsakiri et
al., 2002).
The criteria for the evaluation are presented in Table 5. From
this table it can be indicated that the pan-sharpened image
preserves to a satisfactory degree the spectral information of the
multispectral orthoimage.
Comparison between multispectral — pan-sharpened image
BANDS: ! 2 3 z
Bias (ideal value: 0) 3.16 1.96 1.66 0.73
Correlation coefficient 0.82 0.84 0.85 0.83
(ideal value: 1)
Standard deviation of the 2137 19.59 18.70 16.39
difference image (ideal value: 0)
Comparison between NDVI multispectral —
NDVI pan-sharpened image
188
Bias (ideal value: 0) 2.38
Correlation coefficient 0.91
(ideal value: 1) :
Standard deviation of the
6.40
difference image (ideal value: 0)
Table 5. Criteria on the spectral evaluation of the pan-sharpened
image
4.5 Evaluation of the spatial quality of the pan-sharpened
image
During the production of the pan-sharpened image another
concern is the quality of the spatial information. The pan-
sharpened image should maintain the spatial characteristics of
the initial panchromatic orthoimage and assure this, a certain
assessment of the pan-sharpened image is performed.
The methodology used presupposes the application of a high-
pass filter (Zhou et. al., 1998; Li, 2000). It is known that a high-
pass filter enhances the edges between homogenous groups of
pixels and increases the spatial frequency of an image. In this
case, a 7X7 high-pass filter is applied to the initial
panchromatic orthoimage and the pan-sharpened image. Before
