stanbul 2004
cost is large.
icc in a chip.
echnique in
sk to restore
attributions
ng lightness
However, its
| is lost, and
ieve that the
ncident light
(flect light.
ess through
w frequency
s are the lost
ssion of the
ving retinex
oach [Land,
ı (MSRCR)
y, which can
cy and color
Ovide some
Il aspects. In
1999], it's
iss of data in
, and color
h. However,
is the former
e, literatures
dging effect
> almost the
mputation is
. Which is a
2001] The
f loss some
nto different
equency part
looked as a
While the
e united into
VISRCR can
le, it control
: the unite of
except the
ges in three
cognition, a
ve. The test
algorithm in
amples here
namic range
mprovement
med in, the
Where, L.B.
the intensity
C.C. means
of contrast
rength of the
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
MERE
ds. 2
Fig.5 Homomorphic Filtering
Algorithm LB LIIR.IGL PCC bCI
Simple Template dete : * x
Image Restoration ese x * ko 3
Mask Simulation Ww o * we | x
Homomorphic Filtering | ** idisse: dai Week | xx
MSRCR + xke * Xok kok
*
Tab.1 Compare of 5 main algorithms in 4 indicators
Obviously, each algorithm has its advantages and defects in
processing traditional dodging problem. However, when they
are tested for color images, only the Multi-scale Retinex can
obtain a relative realistic image. The compare is displayed in
Fig. 6 to Fig.9.
Fig.7 Origin
Fig.8 General Dodging Fig.9 MRSCR
The compare indicates that general dodging algorithm can
obtain more excellent traditional dodging effect than MSRCR.
while the MRSCR can obtain more realistic effect than general
dodging algorithm. Additional, if takes the assumption of grey
world [Gasparini, 2004] or makes use of white balancing
algorithm, the effect of color restoration can be better.
Therefore, a framework combines the advantage of general
algorithm, MRSCR and improves its color restoration ability
should meet the extended desire of dodging.
5. ANEW FRAMEWORK FOR EXTENDED DODGING
Based on the analysis above, the task of extended dodging can
be performed by four steps: lightness balancing, color rendition,
color cast elimination, and geometry improvement. Here the
framework proposed is to remove the color cast firstly, and then
unites the geometry improvement and color rendition into the
framework of MSRCR. The last step is to balance the lightness
(See Fig.10). Because color cast elimination will not affect the
lightness distribution and the contrast and geometry quality of
the image, the color cast elimination is performed firstly.
Secondly, because the MSRCR can restore the color and obtain
a moderate dynamic range, it’s processed secondly.
Input image
i
Color cast elimination
J Color renditon
MSRCR » Dynamic compression
v
Lightness balancing
Y
Output image
Geometry improvement
Fig.10 Framework of extended dodging
5.1 Color cast elimination
Because of the changing of color temperature of the illuminant
of the scene, the imaging lens' optical feature, and atmosphere
affection, the obtained image usually represents an obvious
color cast. To obtain a realistic output image, the color cast
should be rectified firstly.
Since the imaging scene consists of complex ground objects in
remote sensing images and aero-borne images, the assumption
of grey world is usually applied. Based on the the Von Kries
hypothesis with the RGB channels considered an approximation
of the L (large), M (middle), S (small) retinal wavebands [Land,
1971]. The estimate of the sensor scaling coefficients is
assimilated within the evaluation of the color balancing
coefficients. The diagonal transform is:
E] [x 0 OR
g'i-lo z. 016 (0
A nt 0 4 18
The gain coefficients, kr, kc, kg are estimated by grey world
assumption:
