+. se €
A c8
ua
V uU tv
vec cuna EN
Ea a a Aa
I:
|
Le |
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
system will be compared GCP image which is generated from
satellite image of the past with the current image.
Figure 4 shows the procedure for the proposed image matching.
GCP images,
Sub-images in search region
F[R Wavelet
im 3
For all
directional bands(m)
» P ME TE
im im
Energy. Analysis
(a) Procedure for Energy Analysis
For all non-overlapped sub-images in
Search region
Energy Analysis
|
Compare(Pom Pin) —
*
Best match with
minimum distortion
(b) Matching Procedure
Figure 4. Image Matching by Wavelet Transform
In the Figure 4-(a), see the equation (1).
TE Y WC, (1)
The symbol WC; denotes wavelet coefficients in Ath band on
the layer j, and if i equals zero, it means chip image, otherwise,
non-overlapped images in search region. Therefore, the symbol
TE is the total energy of wavelet transform domain except the
lowest frequency band. Assume that & denotes the following,
[= band, if 0
k =4 LH band, if 1 (2)
Len band, if 2
The next equation
EQ WC, (3)
im
represents energy of wavelet coefficients as equation (1),
however, it’s not ‘total’, but energy of directional bands. It
differs from ‘4’ because ‘m’ includes all bands on the same
direction. The meaning can be expressed with m = {horizontal =
0, vertical = 1, horizontal = 2}. Although two images are almost
the same, we can not sure that energy distribution of two
images is the same because circumstances of the images such as
brightness, noise, may be not identical. So, the final step in the
analysis is the normalization of energy.
Figure 4-(b) shows the proposed matching process. For the
matching, first of all, search region or reference image must be
decomposed into non-overlapped sub images. And then, energy
of transform domain will be calculated after wavelet transform
is applied to each sub image. From the figure 4-(a) and figure 2,
we know that the variable P;, is the parameter of GCP database.
Final matching step is to compare Py, of GCP with P,, of sub
image.
4. EXPERIMENTAL RESULT
Two images are used to the experimental result, Landsat ETM+
image as reference which was down-linked on September, 1999
and GCP image of October 2000.
(a) Original Image
| (b) Matched Image
Figure 5. The experimental result
Although the chip image is more blurred than matched image of
the original, the simulation result showed good result. This
comes from the property of the proposed method that detects
directional of energy in transform domain.
5. CONCLUSION
This paper proposed the compression method for an efficient
management of satellite image and the image matching scheme.
The proposed image matching can be efficiently used in
application fields that a reference image has a distinct edge
although there is loss in background contents. In the future, the
matching method will be developed so that the matching is
operated on compressed stream immediately.
REFERENCES
Ghassemian, H., 2001. Multispectral image compression by an
on-board scene segmentation, In Proc of IEEE, pp.91-93.
Mittal, M.L., 1999. An efficient and fast compression technique
for multispectral browse images, In Proc. of IEEE, ppl177-
1179.
Park, J.H., 2002. New compression scheme for multispectral
images, Proc. of ISRS, pp. 565-568, 2002
|! (c) Magnified Chip Image
