p
e the
other
'asure
The
theses
a. For
| are
n the
utions
n an
f the
in the
would
itself
ching
ed to
yes in
an be
enting
WS:
h are
ulator
tions;
ration
s are
cation
t-after
eak in
)bable
ak 18
nner),
n, the
t the
n, the
1ieved
cantly
ibuted
ne the
These
Juares
'olved
gested
using
scenes
xhibit
^».
; with
— |o an Na | Na
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
First, the parameters of the registration transformation function
(using 2-D similarity and affine transformation functions) are
estimated using well distributed tie points, which have been
manually identified in the scenes, Table 2. The variance
component (52) in (Pixel ^ derived from the least squares
procedure summarizes the quality of fit between the involved
primitives in the registration process. Smaller variance
component indicates a better fit between the registration
primitives. The selection of common points in the various
scenes proved to be a very difficult and time-consuming task.
Analyzing the results in Table 2, one can see that the estimated
variance component has improved using affine transformation
when compared to that derived through 2-D similarity
transformation.
Table 2. Transformation parameters based on manual point
measurements
Ortho-56 Ortho-72 Ortho-00 Ortho-01
6? 4.3580 2.1334 1.5207 0.8402
a, 95.0619 64.3973 89.8651 52.9031
b, -105.2252 272.1483 73.5173 30.9711
a 0.9164 1.3015 0.3347 0.1587
Eh -0.0185 0.0590 0.0127 -0.0512
Ortho-56 Ortho-72 Ortho-00 Ortho-01
o7 4.1231 1.7976 1.504 0.8089
a, 93.8898 63.4821 90.9108 51.4447
a 0.9120 1.2977 0.3360 0.1572
a 0.0162 -0.0622 -0.0116 0.0493
b, -105.5540 272.0775 73.9394 31.0340
b -0.0216 0.0560 0.0123 -0.0488
b; 0.9196 1.3038 0.3318 0.1601
Afterwards, straight-line segments were manually digitized in
the available scenes. As an example, Figure 4 shows the
digitized segments in Aerial 1956 and Ortho-photo 1999
scenes. In this figure, one can see that there is no complete (i.e.,
one-to-one) correspondence between the digitized primitives in
the input and reference images. The digitized segments are then
incorporated in the MIHT strategy to automatically determine
the correspondence between conjugate line segments as well as
the parameters involved in the registration transformation
function. The estimated registration transformation parameters
as well as the corresponding variance component for all the
datasets are listed in Table 3.
Table 3. Transformation parameters based on
matched lincar features using MIHT
automatically
Ortho-56 Ortho-72 Ortho-00 Ortho-01
à; 2.2298 2.7774 1.7599 0.8977
A 94.0756 65.4424 87.9770 53.1336
b, -106.6365 269.8632 75.8580 30.9736
a 0.9195 1.3041 0.3341 0.1595
bi -0.0210 0.0562 0.0132 -0.0507
Ortho-56 Ortho-72 Ortho-00 Ortho-01
6} 2.1785 2.0657 1.6761 0.8522
ao 94.0991 64.6135 89.5263 52.7716
ai 0.9181 1.3018 0.3355 0.1589
a 0.0181 -0.0592 -0.0105 0.0500
bh. -106.6896 270.2862 75.7333 31.3885
bi -0.0229 0.0542 0.0142 -0.0506
b, 0.9204 1.3053 0.3334 0.1612
931
Similar to the results from the point datasets, the affine
transformation produced better results than the 2-D similarity
transformation. Moreover, comparing the results in tables 2 and
3, one can see that utilizing linear features led to a better fit
between the scenes than that derived using point features. This
should be expected since identifying linear features in multi-
resolution imagery is much more reliable and accurate than
distinct points.
As mentioned earlier, the affine transformation is valid when
assuming relatively flat terrain. In this context, linear features
are advantageous since they restrict the selected primitives
along relatively flat terrain as represented by the road network.
This might not be the case for point primitives that might have
significant relief distortions (e.g., simultaneous considerations
of points along the terrain as well as high rise buildings).
Finally, observing the estimated shift components among the
registered scenes (ao, bo), one can see that the proposed strategy
successfully converged without the need for approximate
registration of these scenes.
= | Ortho-photo 1999 Linear Features
—^ Matched Aerial 1956 Linear Features
- « « Non-Matched Aerial 1956 Linear Features
Figure 4. Established correspondences between the 1956 aerial
image and the 1999 ortho-photo line segments
Figure 4 depicts established correspondences between the
digitized primitives in the Ortho-photo 1999 and Aerial 1956
scenes. The estimated transformation parameters are used to
resample the reference image to the coordinate system
associated with the input image. Figure 5 shows a mosaic image
derived by combining Landsat 2000, Orho-photo 1999, and
Aerial 1956. A closer look to this figure reveals the following
facts:
e Due to the limited area covered by Landsat image 2000,
Figure 5(a), image completion concept has been applied to
obtain full coverage for the city of Calgary. Aerial 1956 and
ortho photo were used to achieve such a task Figure 5(b).
One should note that multi-image integration has been
accomplished. This is an important process that is needed to
cope with large diversity of contemporary available images.
e In Figure 5(c), every other square patch in the reference
image has been replaced by the corresponding resampled
patch in the input image. It can be seen that features (e.g.
roads, rivers, buildings) in the derived mosaic accurately fit
each other (observe the smooth transition along the features
within the resampled patches). This proves the validity of the
estimated parameters of the transformation function relating
these scenes.
