The points that pass these tests will be indicated as reliable
matches. The points that cannot pass these tests may have
multiple solutions. This matching ambiguity will be solved by
the following global image matching method through local
smoothness constraints (see chapter 3.4).
3.3 Edge Matching
To reconstruct a DSM from very high-resolution images over
urban areas we must take into account the problems caused by
surface discontinuities, occlusions and the significant perspective
projection distortion. Even with satellite images, line features are
also important for capturing and modeling terrain features such
as ridgelines and breaklines. Matching the edges is a possible
solution to these problems. However we should consider the
following problems:
e The edge on one image may break into more than one segment
due to image noise, occlusions and the deficiencies of the feature
extraction algorithms.
e The conjugate edges on different images may have quite
different shapes due to the projection distortion.
e There may be many similar features in a search area.
The edge matching procedure presented here is based on the
evaluation of the local geometric and photometric attributes of
edges for the solution of disambiguities. The quasi-epipolar
geometry and the DSM data derived from the higher-level of the
image pyramid are used to provide for the matching candidates
for each edgel. A figural continuity constraint satisfaction
scheme (the disparity along an edge should change smoothly)
and a shape matching approach are used to achieve the final
results.
The well-known Canny operator is used to locate the intensity
discontinuities. Then edgels are linked into free-form edges
through a local processing that analyses the characteristics of
these pixels in a small neighborhood. This approach is carried
out independently on three images. Only edges above a
minimum length (30 pixels for SI images and 15 pixels for
satellite images) are considered for matching.
The edgels along the given edge are matched with the edges that
are defined at the intersection points between the candidate
edges and the correspondend epipolar curve within the search
window on one of the search images. The search window can be
determined by using the same method as in chapter 3.2. There
may be several matching candidates within the search window.
To solve this ambiguity problem we perform the following three
steps sequentially:
a) Evaluation of the difference of the local edge orientation
between the given edgel and its candidates. The local
orientation for an edgel is the image intensity gradient
computed modulo 27. Candidates with differences above a
threshold will be dropped. Considering the possible relief
distortion, this constraint should not be too tight. For
example, we use 40 degrees for SI images.
b) Evaluation of the normalized cross-correlation coefficient
of the intensity values on each side of the edgel. Exclusion
of those candidates that have a very low correlation
coefficient, e.g. less than 0.5.
c) If more than two images are available, each candidate can
be validated on the third image through the indicator used
in step b).
After these three steps, the given edgel may still have more than
one candidate. The problem will be further solved by using the
figural continuity constraint through a relaxation method. This
method examines the candidates by computing how much
support they receive from their local neighborhood along the
edge. We select the candidate that gains the highest support as
the correct match for each edgel. For each edge, the edgels that
have only one candidate will serve as “anchors” for this
relaxation method. By linking the successfully matched edgels in
the search images, we obtain the correspondent edge matc.
Finally, we do a shape matching between the given edge and its
correspondent edge through least squares adjustment. Only the
edges with small shape matching errors will be kept.
Figure 3 shows an example of our edge matching.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Figure 3: Edge matching with SI images (5 cm resolution)
3.4 Grid Point Matching
We use grid points to create uniformly distributed points over
the whole images even in areas with very little or no texture. The
correspondences of these grid points could be computed by using
the method presented in chapter 3.2. Compared to the feature
points, the choice of grid points is blind and thus many grid
points are located in areas with weak or no texture. The search
for the match of a given grid point has more possibilities to yield
multiple candidates, or even no candidate.
To solve this problem, we use a global image matching method
with relaxation technique. This method examines the candidates
by computation how much support they receive from their local
neighborhood and select the candidate that gains the highest
support as the correct match. Here we use Prazdny's "coherence
principle" model (Prazdny, 1985). We incorporate this idea in
our global image matching and get the solution by relaxation
technique.
Firstly, the points are selected in form of a regular grid in the
reference image. Their matching candidates on the search images
are computed. Together with al! the matched feature points and
edges they construct a TIN. It should be noted here that all the
matched points can be categorized into three classes: Points
having reliable matches, points having several candidates, and
points without matching candidates. In the first case, they are
treated as having only one matching candidate and, they serve as
“anchors” for the global matching procedure. For the last case,
they will be given several “false” candidates (with a very small
correlation coefficient value) evenly distributed within the search
window. The matched edges serve as break-lines in the TIN
structure. They control the weights of the local smoothness
constraints.
This method is performed on stereo pairs. The key point of this
method, that distinguishes it from the single point matching, IS
its compatible coefficient function and its smoothness constraint
satisfaction scheme. With the smoothness constraint, areas with
homogeneous or only little texture can be bridged over,
assuming that the terrain surface varies smoothly over the area.
In the meantime, the surface discontinuities can be preserved
because the smoothness constraints cannot cross the edges. For
details of this procedure see (Gruen, Zhang, 2003).
3.5 Matching Through the Image Pyramids
A triangular irregular network (TIN) based DSM is constructed
from the matched features on each level of the pyramid, which in
turn is used in the subsequent pyramid level for the
approximations and adaptive computation of the matching
parameters. The matched edges are used as breaklines such that
no triangle crosses these edges. The TIN maintains the original
matching results without any interpolation. The surface
discontinuities of the terrain can be well captured and preserved.
The initial DSM for the highest level of image pyramid can be
extracted by standard cross-correlation based on a “region
growing” matching strategy. This method uses the already
measured control and tie points as seed points and matches the
points under the assumption that points in a local neighborhood
should have similar disparities (Otto, Chau, 1988). This method
is justified because the disparity surface can be treated as
continuous and smooth on the lowest resolution image pyramid
level. In some difficult areas like very rough alpine terrain, some
International À
es
manually meas
It will give bett
3.6 Refined M.
MPGC (Multi-
developed by
Baltsavias, 19
geometric con:
space and peri
object coordina
be used simult
pixel range.
Our modified a
integrates the
Array sensor
matching to sec
around the epi
matches
| Figure 4
Top:
The modified |
order to achiev
approaches (3.:
the MPGC pro
initial values
predetermined
DSM data. Th
matching wind:
derived DSM :
Through ima
coordinates in
initial values o
from these four
way, features c
multiple image
image scales. I
strips.
For edge match
object edge ar
parameters of
Some points, e:
texture areas, v
kept but they ar
In order to eva
generation it h
textures, terrain
report about D
SPOTS HRS in
4.1 SI Image D
In Japan's GSI
images and aer
on the compari
aerial photos ai
Images.
