tion 
d as 
the 
lef ined 
an 
» 
7 
-ithm 
3. of 
the 
ment 
One 
[g a 
to 
The 
neas- 
4 
, and 
image 
neas- 
(rack 
mean 
vidth 
the 
~~ 
-— 
e 
~~ 
~ 
oo 
> 
T 
  
L| 
As ère 
The coarsest image computed crack pattern 
in 5 cm meshes 
computed items 
Total length of crack : 210.64 cm 
Density of crack : 842.56 cm/n* 
Mean of crack width : 0.29 mm 
Fig. 12. The coarsest image examined and its outcome 
( sample No. 9 ) 
5. DISCUSSION 
In the present examination 34 % of cracks 
are undetected. It is realized that most 
of undetected cracks are little contrast 
to the concrete due to a certain stains 
around them. The line-edge filter should 
be more sensitive in such a condition 
consequently. In order to make the filter 
sensitive, thresholding in edge detecting 
algorithm needs to be improved. In 
particular a method of calculating 
threshold value should be changed  suit- 
ably according to a condition around the 
crack. 
The measurement of crack width has been 
accomplished with accuracy of the mean of 
RMS error 0.08 mm, which is attained the 
goal of this study intended. In order to 
achieve more precise measurement of crack 
width, it is needed that a resolution of 
the original image should be more fine. 
6 . CONCLUSION 
To assess the deterioration of concrete 
structures, an algorithm employed in a 
prototype of crack measurement system 
based on hierarchical image processing 
technique has been examined with 18 
samples of concrete crack images. The 
algorithm .can be executed to detect and 
measure the cracks ranging from 0.1 to 
3.0 mm in width, in which a series of 
images are generated and structured 
hierarchically from fine to coarser, and 
a single spatial filter is used for crack 
detection. 
As a result, 66 % of existing cracks on 
the samples have been detected and meas- 
urement of crack width has been performed 
with an accuracy of 0.08 mm RMS. It is 
clear that the edge detecting algorithm 
should be improved to prevent from misde- 
tection of cracks which are little  con- 
trast to concrete due to a certain 
stains. The condition around cracks, 
however, are changed so much in places 
that no useful algorithm can be prepared 
practically to apply to detecting every 
crack influenced by every stain. 
It is concluded that the algorithm  exam- 
ined in this study is useful to the crack 
measurement system, although a  improve- 
ment of algorithm should be necessary to 
eliminate such misdetection. 
References 
Hilditeh, C.J., 1969. Linear skeletons 
from square cupboards. In: Machine Intel- 
ligence IV, Meltzer, B., Michie, D., 
Eds., University Press, Edinburgh, pp. 
403-420. 
Ohtsu, N., 1980. An automatic Threshold 
Selection Method Based on Discriminating 
and Least Squares Criteria. IEIC Trans., 
J63D (4) : 349-356. 
Suzuki, H., Ito, S., Suzuki, A., Mori, 
M. 1985. Chest X-ray Image Analysis 
System to Detect Lung Cancer. In: 16th 
Conference of Image Technology, pp. 145- 
149.