Small isolated holes
in a thresholded area
Fig.9. Noise reduction by filling small holes
3.4 Detailed Positioning
Each vector of cracks must be centered in
thresholded area, as the location of
mapped vectors have ambiguity. Fig. 10
shows a schema of positioning process.
In this processing location of a crack
vector is shifted. into the midpoint
between both sides of intersection which
are determined by the cross section at
right angle of crack direction. Further-
more, the vector mapped on an isolated
area is moved to the area on which the
majority of vectors have been mapped.
» WN
|. An edge of a crack area
> Direction of crack
A : An intersection point on
an edge of a crack area
B : The initially mapped point
of a crack vector
C : The midpoint between A
and B
D : An intersection point on
an edge of a crack area
Fig. 10. À schema of detailed positioning by centering
3.5 Measurement of crack width
Crack width are measured at the location
of each crack vector, which is defined as
a diameter of an inscribed circle of the
thresholded area shown in Fig. 11.
Crack width is defined
as a diameter of an
inscribed circle.
Fig. 11. A definition of crack width
4. THE TEST OF THE ALGORITHM
Hierarchical image processing algorithm
has been examined with 18 samples of
concrete crack images. In the test the
resolution of finest image is 0.025 mm
per a pixel. Results of the measurement
can be evaluated by the two indices. One
is an index of extraction which is a
ratio the length of extracted cracks to
the total length of existing cracks. The
other is RMS error in crack width meas-
urement.
Table 1 shows the summary of results, and
Fig. 12 shows one of the coarsest image
examined and its outcome of crack meas-
urement. The result shows that the crack
recognition has been achieved in the mean
of extraction ratio 66 % and the width
measurement done with accuracy of the
mean of RMS error 0.08 mm.
Table. | The summary of results of the test
d ui m e LL 9 rt oe n RMS error
(cm) (X) rm
1 38.63 31.00 80.2| 0.11 (20
2 41. 47 23. 12 55.8| 0.07 (20
3 25. 65 20. 49 19.9| 0.06 (20)
4 183. 27 103. 77 56.6 | 0.14 (20)
5 209. 99 123. 50 58.8 | 0.04 (19)
6 301. 75 245. 90 81.51 0.17 (38)
7 290. 82 67.83 23.31 0.05 (20)
8 156. 24 105. 75 67.71 0.05 (19)
9 428. 60 210. 64 49.1 | 0.06 (33)
10 71.22 51. 78 72.7| 0.04 (16)
11 63. 06 53. 50 84.8| 0.08 (20)
12 55. 07 38. 99 70.8 | 0.05 (20)
13 30. 36 27.81 91.6| 0.11 (20
14 46.87 40. 16 85.7 | 0.05 (20)
15 31.21 20. 22 64.7| 0.08 (20)
16 25. 16 19. 29 76.7| 0.09 (20)
17 39. 75 22. 02 55.4| 0.06 (20)
18 31. 33 12. 29 39.2| 0.05 (20)
mean valuc 66.4| 0.08
X) Number in () indicates the number of samples.
