nbul 2004
a) x)
tained
result
r view
nds of
. Mars
(Left
| Sol 9
:18:01
efects,
rent to
itional
e best
4) for
nation
xdel is
m 3.2.
age to
mbine
age. In
ted to
panel.
wc can
image
cus as
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
Where g is the degrade image, g; is image on focal panel, g; is
the i bit panel image, & denotes the ‘and’ operation.
Once /(x,y) is determined, the image restoration can be easily
done to r(x,v) in wavelet domain. There are many restoration
algorithms in wavelet domain (M.R. Banham, 1996, M. Belge.
2000, Y. Wang, 1999, M. Crouse, 1998, R. Chan, 2003).
with geometric correction before mosaic. Thus, the mosaic slot
is missed after this processing. We can see obviously in the 1:1
zoom windows as figure 11(c) and figure 11(d) that the mosaic
slot of buildings can be effective removed. Figure 12 shows
result of a mosaic project with 8 IKNOS images.
No
Bit
Synthesis
Bit
Slicing
8 Gy)
Exp fosy)
r(x,y) !
| { i
||
Wavelet j deem
Transform DL:
———M
i r X,Y,
—] | Sobel Edge | Edge o — i
» I
—7 — 1 Detection
nverse
Coefficien :
Wavelet
Figure 9 Restoration Based On RW Model
Considering the blur edge degradation, an edge preservation
filtering to wavelet coefficients with the edge detection is
proposed in this paper to do image restoration in RW domain.
Figure 9 gives the total processing flow.
Sobel operator is used for edge detection to the wavelet
coefficients to make edge clear and non-edge coefficients are
taken as noise and removed. Furthermore, the regularized
image restoration based on common imaging model as figure 1
shown can be taken to i(x,v) if large sensor movement exist
during imaging. Figure 11 gives the restoration result based on
RW model to SPOT image.
The focal panel image is composite of 3, 4, 6, 7 and 8 bits while
the non-focus image is composite of 1, 2 and 5 bit. Bit 1 and bit
2 images are almost noise, while the bit 5 image is the contour
image. As figure 10 (a) and (b) shown, the blur edge becomes
clearer. Table 1 gives the quality assessment result of Wiener
Restoration, Inverse Filtering and RW based restoration
proposed in this paper.
Table | Image Restoration Comparison
Wiener Inverse | RW Based
Restoration | Filtering | Restoration
Mean Square Gradient 5.148 5.800 9.214
Definition 0.21 0.24 0.45
As table | shown, the restoration result based on RW
imaging model is superior to wiener restoration and
wiener filtering based on the traditional imaging model as
figure 1 shown.
3. EXPERIMENTS AND RESULTS
For photogrammetry and remote sensing, image mosaic is very
important to produce orthodox image map. It is very common
that the adjacent two images or adjacent two stripe image have
different contrast and tone, which will make image mosaic fail
with obvious mosaic strip as figure 12(a) shown and ‘colour
cloth” phenomena to high resolution satellite image as figure
13(a) shown.
In order to remove the mosaic stripe. image dodging based on
RW imaging model to the adjacent two images is done together
965
4. CONCLUSIONS AND DISCUSSIONS
Retinex Wavelet imaging model based provides a
reasonable image decomposition and representation,
which is different to traditional imaging model. This
model distinguish image with ill-problem part and non-ill part
to avoid image quality improvement processing together
damaging non-ill part and fully consider the characteristic of
HSV. The image dodging enhancement to different kinds of
image defects show that it can effectively deal with non-even
light, contrast and tone problems. The image restoration based
on RW model to blur edge is better than common restoration
algorithms. The typical application of image mosaic shows the
advantages of proposed RW imaging model in the field of
image quality related applications. And also it can improve the
image quality such as shadow defects to restore objects, which
is very useful to image understanding processing such as object
identification, image matching and image vision. Most
important, the retinex wavelet based image representation
model is a simulation of HSV and its sensitive course, which
can be used for image quality assessment system. The authors
think the quality assessment system based on RW imaging
model can combine quantitative assessment and personal
assessment standard and it will be a research direction. of
uniting quality assessment system. Still some other problems
and researches in this work to be done in the future are listed as
following items:
(1) More imaging processing algorithms based on rctinex
wavelet model such as image fusion, image compression
and image understanding.
(2) Effective and accurate estimation of /(x.Y)
Image quality assessment based on retinex wavelet
model
ACKNOWLEDGEMENT
The author would like to thank Wuhan Supresoft Co., Ltd to
provide some experiment data and to provide IKNOS test
images and NASA Mars website www.marsrover.jpl.nasa.gov
to provide Mars images.
