IARS
they have
SA Mars
ment and
ie lessons
shows the
se craters
/ unlikely
utomated
yS-MOC)
ater types
planetary
om three
(2) inter-
idealised
rs and the
03).
ave been
a mining
ality for
999). No
Xf impact
relatively
y aim of
rater and
eveloped
| Hough
esult was
ft vector
ted using
utomated
template
ar crater
DTM and
1. The
ments in
focusing
Matrix)
These
| optimal
yr impact
d using à
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
Hough transformation. They are then finally verified using
template matching.
In the case of small impact craters, which are classified by
measuring their ROIs' size, the Rols are directly fitted to an
optimal ellipse without any further processing.
- quwetobeo o a nme
aw image De
I
|
p Selection of RO! (Region of interest) i
C eLeM 7 and local edge detection i
|
i
!
= 1
Edge c
1 Li
[nF zi 3 Preliminary crater edge selection by |. :
angle I edge direction analysis T 77
|
i
BEE sna =a E EE riri
l'Organisation sage i 1
Formation of primitive arcs |
il
| Organization of Arcs |
been ln a e menie mete m oim
1
-——— — — — — bPQ——-——————
eo)
|
IF size of ROI
* threshold
Primitive ellipse formation and
fitness check
“
i
t
i
| !
I
I
t
| | Rim refinement Le
I 1 1
I il i
Predefined |. |. |
template T
!
Meu mm — — — — — — — — — — — — — —
1 Verification by template |
Figure |. Overall work flow and processing steps
2.1 Focusing
When applying the first stage, there are usuallye too many
connected edge segments in an optical image. For example,
hundreds of edges appear in a single MOC image even on a
relatively flat area. Therefore it is impossible to apply these
algorithms to these edges to drive crater shape. A focusing
strategy using the GLCM and edge direction analysis is used
here to reduce the search space.
At first, an edge is localized by it's ROI, which is defined
through GLCM texture classification. Then within a localized
edge area, edge thresholding is applied.
However, extracted edges usually include not only crater rims
but also shadow boundaries. As seen in Figure 3, which is
generated from a generalised 3D crater using a Phong shading
model, four different shading regions can be defined and the
boundaries of each shading form double structured edge lines.
The real crater edge is usually the boundary between the
illuminated and shadowed areas.
Widéden eds Time
ntt
Figure 2. Edge formation geometry in specific illumination
condition
817
3
(a) Simulated image using
hill shading for a crater DTM
model using a Phong shading
model Solar Evevation 745^,
Solar Azimuth=0°)
(b) Detected edges from
image (a)
Figure 3. Simulated crater image and edges
The analysis of edges in these four regions by looking at the
directional properties shows that the centre point of the crater
rim part should satisfy the following condition.
= (1)
+ 3
where 6 = sun azimuth angle,
Q-edge direction of centre point
Additionally the extent of one crater edge rim is limited by
P.-1WM2<HP<O,+ M2 (2)
where o , — edge direction of centre point ,
9 —edge direction of crater rim
assuming that there are no hidden edge lines from erosion or
other illumination effects.
The detection of preliminary crater edge can be simply
implemented by rotating the edge mask or a algorithms
discussed below.
At first, the marginal degree of the central peak is defined in
individual ROIs by (3),
De 2m (3)
" 0.5max(Dx, Dy)
where Dx = X dimension,
Dy =Y dimension of connected component
so that if the thresholded part includes edge segment, which
satisfy (4)
Q.*9 «99.94 (4)
where q ,. edge direction of centre point.
Then we can define this as a preliminary crater rim edge.
Then re-arranging all of the edge pixels with r (estimated centre
from initial conic fitting) and « space after finding the
maximum intensity point in each « , interval to detect seed
points. By applying region growing with these seed points using