)n matrix
e point co-
ersections.
nder appli-
h accuracy
nuous sur-
s useful to
at purpose
Digital Ele-
tal Surface
s standard
| elevation
so enables
erspective
e Ebner et
r of a
probe
1 point co-
)eaks. The
| matching
suitable for
to remove
lue for the
ical control
with Data
ror of the
a working
'estrictions
Yoved from
In case of the feature-based processing approach, the
derived nodes and edges have to be converted into a
CAD data format with the basis elements of points and
lines. This converted data are available for the visuali-
zation and further processing with CAD systems.
4. EVALUATION SAMPLES
4.1 Reconstruction of the Surface of a Copper Sample
The aim of this investigation was the determination of the
topography of dissipative chemical patterns spontaneous-
ly developed during etching procedure of thin copper
films. The copper sample shown in Fig. 7 has been used
in a SEM (Zeiss DSM 960) for the acquisition of an image
series with tilt angle steps of 5?, at an acceleration voltage
of 30 kV and a magnification of 3000.
Ë d i BiH
Fe
t
Figure 7: Copper Sample
(SEM Image with a Magnification of 3000:1)
For the subsequent processing of the images, realized on
a SGI-Workstation, two images with tilt angles of.- 5? and
+ 5° were selected. Ten points were measured in an inter-
active way in both images for the orientation process. The
difference between the scales in x- and y-direction has
been determined by a previous calibration of the system.
20 p
Figure 8: Surface Points in x/y-Plane estimated with
Least-Squares Matching (Dimensions in pm)
After the successful estimation of the orientation parame-
ters follows the automatic correlation process. The area-
based least-squares matching yields a success rate of
nearly 80%. Fig. 8 shows the results of the matching pro-
cess with a threshold for the correlation coefficient at 0,8.
The next step is the estimation of coordinates in object
space. The obtained three-dimensional cluster with nearly
75000 points (Fig. 5) was used for the generation of a
Digital Surface Model and for all following operations and
visualization. Fig. 6 shows a perspective view of the DSM
data. Finally — as seen in Fig. 9 — a shading map, gene-
rated with the DEM software, has been mapped over a
perspective grid model.
Figure 9: Shading Model of Copper Microprobe
4.2 Geometrical Reconstruction of a Silicon Sample
The micromechanical structured silicon sample (see Fig.
10) is provided with typical features of microstructures:
Strong edges and surfaces with poor texture. For this
reason we chose this sample to test the possibilities of a
widely automated reconstruction of the geometrical shape
of such probes.
B m | uu (17
i i NN D SENS Mesh du
D
|
N
Figure 10: Silicon Sample
(SEM Image with a Magnification of 500:1)
As already described, the automatic feature-based pro-
cessing approach includes a number of operations. The
first step is the detection of edges in one reference image.
For this purpose we compare four different approaches:
— Canny-Operator
— Deriche-Operator
— VDRF-Operator (from Khoros)
— SUSAN (Smallest Univalue Segment Assimilating
Nucleus)
The best results, achieved by the Canny-Operator, are
shown in Fig. 11. For more information about this special
operator see Canny (1986). The result of the edge ex-
traction is a binary image with extracted edge pixels. To
229
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
