International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Figure 2: Left: The Canny output map. Right: The Rothwell
edge map.
The input parameters to the /verson-Zucker algorithm were the
following: a) Threshold (7), number of directions (d) for the
algorithm to follow (1-16), degrees of freedom based upon the
number of directions (4-64), the output detection type (E
(edges), P (Positive lines), and N (Negative lines)). The best
output result (0.4967) (Table 1) was derived using 16 directions
and 7=0.015.
The input parameter to the Black algorithm was the smoothing
coefficient (c) in the range (0, 1]. The best output edge map
resulted in the highest Pratt metric (0.4773) (Table 1) using
0=0.25 (Figure 3).
Figure 3: Left: The Iverson-Zucker output map. Right: The
Black edge map.
The input parameters to the SUSAN algorithm by Smith and
Brady were: (a) Brightness threshold (-7) (default = 20), (b)
distance threshold (-d) (default = 4.00) (used instead of flat 3x3
mask), (c) use of flat 3X3 mask (-3), (d) choice among edges (-
€), smoothing (-s) or corner detection (-c) modes. The best
output result (0.3171) (Table 1) was derived for 7715.00
(Figure 4).
Figure 4: The SUSAN output map. The edges are depicted
with black lines and overlayed on the smoothed image.
792
Using the EDISON algorithm, the best output edge map
(0.4507) (Table 1) was derived using the following combination
of parameters: a) Gradient=2.00, b) Minimum length=5.00, c) —
e) Nonmaxima suppression: Type=arc, Rank-0.5 and
Confidence-0.4, f) — h) High Threshold for hysteresis: Type —
box, Rank=0.91 and Confidence=0.92, and finally, i) — k) Low
Threshold for hysteresis: Type=arc, Rank-0.98 and
Confidence-0.93.
Finally, the parameters used in the Bezdek algorithm are: (1)
Tau, in the range [0.0...5.0], (2) Chi, as a function
HTau)=2.0*Tau, (3) Gamma, as f(Tau)=2.0*Tau, (4) Omega, as
function of f{Tau)=3.0*Tau, (5) Binary Threshold, which is in
the range [0...GRAY LEVELS — 1] and (6) Edge Features
(Sobel). The best result (0.4428) (Table 1) was obtained for
Tau=1.00 and Binary Threshold=80.00 (Figure 5).
Figure 5: Left: The EDISON edge map. Right: The Bezdek
edge map.
In a similar manner and logic, the selected edge detection
algorithms were further applied on the DEM of the same area,
and, only two of the best results as judged by
photointerpretation (Table 1) are presented in Figure 6 due to
paper size constraints.
AA NN
^ e ec
EC. © J / ie
d UN NT es
{ | “A
Cos j& f ‘ X
CC TN yr Em
Figure 6: (a) Based on DEM processing: the Canny edge map
(left) and the EDISON edge map (right).
In the Canny algorithm, the parameter set with the highest score
of the Pratt evaluation metric (0.4332) (Table 1) was for
o=1.50, Tlow=0.30 and Thigh=0.70.
In the EDISON algorithm, the best output edge map (0.4359)
(Table 1) was derived using the following combination of
parameters: a) Gradient=2.00, b) Minimum length=4.00, c) — e)
Nonmaxima suppression: Type=arc, Rank-0.5 and
Confidence=0.7, f) — h) High Threshold for hysteresis: Type =
box, Rank=0.93 and Confidence=0.96, and finally, i) — k) Low
Threshold for hysteresis: Type-arc, Rank=0.97 and
Confidence=0.93 (Figure 6).
—
ea pep £F ^ M
quA 0 e 9 C 1n978
In
d
al
Cc