ul 2004
(4)
(6)
equation
d 1. It is
e surface
constant
ick wards
oped.
(7)
n cosine
Minnaert
'ographic
oughness
| of over-
solved.
| the two-
onsists of
models,
the time
data. This
ovided by
ded relief
ch of the
to derive
.
(8)
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
where SDN = the normalized radiance data for pixel(i,j)
in band(A)
DN ,. = the raw radiance data for pixel(i,j) in band(X)
4, = the mean value for the entire scaled shaded
relief model (0,255)
X, = the scaled (0,255) illumination value for
pixel(i,j)
C, = the correction coefficient for band())
In order to calculate the calibration coefficient, the spectral
responses from large samples falling on the slope facing to and
away from the sun need to be taken. The correction coefficient
can be calculated using equation (9).
i (s, -Ni) (9)
( N, «nn - (s, T
: Hy F A
where N , = the mean on the slope facing away the sun in
the uncalibrated data for the forest category
S, — the mean on the slope facing to the sun in the
uncalibrated data for the forest category
y, the mean value for the entire scaled shaded
relief model (0,255)
p the mean of the illumination of forest on the
slope facing away from the sun.
HT the mean of the illumination of forest on the
slope facing to the sun.
As the main cover type on the shady slope is Forest, it is
chosen to demonstrate calculation of the calibration coefficient.
However, the calibration coefficient calculated will then
perform best in this cover type.
4. RESULT AND DISCUSSION
4.1 Visual Analysis
As the visual effect is more impressive in the near-infrared
band than in a visible band, a visual comparison of the false
colour images is carried out. Figure 1 presents the false colour
IKONOS images of, respectively, (a) the original, (b) cosine
corrected, (c) Minnaert Corrected and (d) normalized images.
By comparing the images visually, we can see that figure 1(b)
has the problem of overcorrection. The corrected image
appears to be too bright in both the sun facing and sun-shaded
slopes. From figure 1(c), the Minnaert corrected image, it is
found that the visual effect on the slope facing away from the
sun has improved significantly. Both the slope facing to the sun
and slope facing away from the sun with same cover type
appear to have similar brightness. However, it should be noted
that the problem of cast-shadow, where the incident angle i >
90 degrees, is not handled in this correction method and
remains to be uncorrected. From figure 1(d), it can be seen that
643
there is less improvement of brightness compared to figure |(c).
However, the increase in brightness of the slope facing away
from the sun can still be seen.
4.2 Statistical Analysis
In order to test the accuracy of the topographic corrections
statistically, training areas of forest and herbaceous covers
were created on both sunny and shady slopes of the images.
Mean values and standard deviations of the training areas for
each band before and after the corrections were calculated and
compared. Tables 1, 2, 3 show the mean values and standard
deviations of the entire study area, herbaceous cover area and
forest cover area respectively. If the correction is successful,
the mean values of each cover types on the shady side should
increase while those on the sunny side should decrease. For the
values of standard deviations, if the correction is successful,
the value of each class should decrease. Table 4 shows the DN
mean value of different cover types on sunny slope and shady
slope of band 4 before and after the topographic correction.
Mean values on the sunny slopes are expected to decrease
while those on the shady side are expected to increase for a
successful topographic correction.
Tables 1, 2 and 3 show that the cosine correction gives the
worst result. There are significant decreases of the mean values
together with large increases in the standard deviations in the
entire image, herbaceous area and forest area of this image
when compared to the original. From table 4, we can also see
that the image is overcorrected as the mean values on the shady
side become much higher than that on the sunny side. This
supports and confirmes the general observation of
overcorrection of previous researchers e.g. Jones ef a/. 1988.
Surprisingly, the Minnaert corrected image, which gives the
best appearance in visual effect, shows a decrease in the mean
values and increase in the standard deviations among the three
classes in bands 1, 2 and 3. But the extent is relatively small
when compared to the cosine corrected image. However, band
4, shows a decrease in the standard deviations for all three
classes. Moreover, from table 4, the mean DN values on sunny
and shady slopes are similar after the corrections. As a result,
this minnaert correction can also be defined as a successful
correction.
The normalization method preserved the mean values of the
entire image and reduced the values of standard deviations. In
the herbaceous area, the mean values decreased slightly and the
standard deviations decreased in all the bands. In the forest
area, the mean values increased in all four bands and the
standard deviations decreased. The decrease of standard
deviation is a clearer test of accuracy of each class. In table 4,
it also supports the finding, the differences of mean values on
sunny slopes and shady slopes reduced after the correction.