Measurement Science and Technology

You may also like

- Experiments on Distributions of Cycle
Measurement of the degraded depth in Degradations of Li-Ion Batteries
Daisuke Sasaki, Keita Hiyoshi, Shuji
cementitious materials by automatic digital image Tsukiyama et al.

- Postnatal morphogenesis of the endocrine

processing part of pancreas of nutria (Myocastor
coypus)
S Dannikov, A Kvochko and O Dilekova
To cite this article: I Segura et al 2010 Meas. Sci. Technol. 21 055103
- Analysis of rainfall data to develop the
classification criteria for identifying agro-
climate resources on the field
A Pramudia, E Susanti and M Auranda

View the article online for updates and enhancements.

This content was downloaded from IP address 144.214.9.191 on 15/05/2023 at 12:14

IOP PUBLISHING MEASUREMENT SCIENCE AND TECHNOLOGY
Meas. Sci. Technol. 21 (2010) 055103 (12pp) doi:10.1088/0957-0233/21/5/055103

Measurement of the degraded depth in

cementitious materials by automatic
digital image processing
I Segura1,4 , M Molero2 , S Aparicio2 and A Moragues3
1
Mechanical Engineering Division, Cartif Technology Centre, Parque Tecnológico de Boecillo,
Valladolid 47151, Spain
2
CAEND (CSIC-UPM), Arganda del Rey, Madrid 28500, Spain
3
Department of Civil Engineering: Construction, School of Civil Engineering of Madrid, Polytechnic
University of Madrid, 28040 Madrid, Spain
E-mail: ignseg@cartif.es, molero@iai.csic.es, sofia.aparicio@iai.csic.es and
amoragues@caminos.upm.es

Received 30 October 2009, in final form 22 January 2010

Published 26 March 2010
Online at stacks.iop.org/MST/21/055103

Abstract
The combination of a staining method and an automatic digital image-processing algorithm is
presented here, to measure degradation depths in cementitious materials. The measurement of
those degraded depths is usually made by direct visual measurements, resulting in many errors
and low reproducibility. The automatic digital image analysis (ADIA) method proposed here
is mainly based on the differentiation of the degraded zone and the sound zone on the basis of
the image histogram. The method comprises several steps, such as sample alignment, image
calibration, background subtraction, image filtering, automatic segmentation and final
measurement of the degraded depth. The algorithm developed has been used to measure the
degraded depths of a set of decalcified cement mortars, made from different cement types and
with varying w/b ratios. Relative to previous methods, this automatic procedure improves the
precision (about 0.03 mm) and the statistical representation of the measurements. The results
obtained by ADIA were compared with direct visual measurements with a very good
correlation (R2 = 0.96) and a mean error of 6%.

Keywords: characterization, image analysis, degredation, cement, phenolphthalein test

(Some figures in this article are in colour only in the electronic version)

1. Introduction microcracks). Some examples of the use of staining techniques

in cementitious materials to reveal microstructural features and
Several degradation processes affecting cementitious materials the effects of degradation processes are, namely, identification
imply modifications of the microstructural characteristics
of the alkali–silica reaction [1], quantification of air voids [2],
of the material, e.g. decrease in pore solution alkalinity,
decalcification of the cement matrix, carbonation of calcium- differentiation of water-to-cement ratios [3], quantification
hydrated compounds and formation of new compounds. of microcracks [4], identification and quantification of both
All these modifications in the cementitious matrix may be carbonation [5] and decalcification processes [6]. Most of
revealed by applying some kind of stain to the sample’s these degradation processes involve a degradation front that
surface. Staining also serves to detect other microstructural spreads inwards the cementitious materials. An example of
features of the cementitious sample (e.g. air voids, porosity or
the application of these simple staining techniques to follow
4 Author to whom any correspondence should be addressed. degradation front progress is the phenolphthalein test. This

0957-0233/10/055103+12$30.00 1 © 2010 IOP Publishing Ltd Printed in the UK & the USA
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

Table 1. Dosage of cement mortars used in this work. 4, 11, 21 and 32 days. This immersion method accelerates
Sample the decalcification process 150 times compared with the same
process under deionized water but diffusion remains the rate-
Component (g) MC4030 MC5030 MC6030
controlling factor [23]. Further information about the entire
Cement 450 450 450 experimental procedure can be found in [24]. Degraded
Sand 1350 1350 1350 samples were cut in slices of 5 mm thickness with a Secotom
Water 180 225 270 10 precision cut-off machine. Subsequently, the samples were
Water/binder ratio 0.4 0.5 0.6
Sand/binder ratio 3/1 3/1 3/1 sprayed with the phenolphthalein standard solution [5, 7] and
images were obtained from the samples’ surfaces.

test is widely used in cementitious materials to measure

carbonation depths [5, 7–11] and decalcification depths [12]. 3. Developed methodology for automatic digital
It is possible, by staining the samples and obtaining image analysis
digital images of the samples’ surface, to measure the
progression of the degradation front in a straightforward, A scheme of the procedure developed to obtain degradation
precise and reproducible way by using automatic digital image depths by using image analysis can be seen in figure 1. It
analysis (ADIA). Although ADIA is widely used for the consists of four different steps: image acquisition, image pre-
analysis and quantification of several microstructural features processing, image processing and extraction of parameters
in cementitious materials [2, 13–17], measurements of the from the image. Each step will be described thoroughly in
‘degraded’ zones (carbonated or decalcified) as shown by the next sections. Although the first two groups of operations
the staining are still done by direct manual measurements are not properly devoted to image analysis, they are quite
[18, 19]. This kind of measurement led to many errors important to obtain high-quality images. This allows us to
and low reproducibility. Those errors are mainly related to get precise and reliable results. The algorithm developed
failures in the visual judgement, to the equipment used for the for acquiring, pre-processing, processing and analysing
measurements and to the fact that it is time consuming and the images was developed under the numerical computing
expensive to have several measurements from a given image. environment and programming language MatlabTM .
Thus, it is necessary to develop ADIA methodologies that
allow precise and reproducible measurement of the degraded
depth. ADIA is easily applied to this problem because the stain 3.1. Image acquisition
displays colour differences between each zone in the sample’s The image acquisition process must be done in such a way
surface. An application of the methodology described below that ensures the highest quality of the images to be analysed.
was presented by the authors in previous papers [20, 21]. Several parameters are controlled at this stage, i.e. sample
This paper describes in detail the methodology developed to illumination, image background and image acquisition device.
determine the degraded depths in decalcified cement mortars
All these parameters will determine both the quality of the
by means of ADIA.
image and the quality and precision of the further analysis. In
this case, those parameters were fixed to obtain a simple but
2. Materials and experimental methods precise and reliable measurement procedure.
As for the image acquisition device, a digital compact
Prismatic mortar bars were made according to current
camera was used for sample surface digitalizing. Since the aim
standards [22] as shown in table 1. Five different commercial
of this work was to develop a simple and easy methodology,
cement types were used: a Portland type I cement with sulfate-
commercial devices were used. The digital compact camera
resistant characteristics (CEM I), two Portland-composite
was an Olympus C-765 ultra-zoom camera with macro mode,
type II cements with fly-ash additions (CEM II/A and
which uses a 1/2.5 inch 4 million pixel CCD sensor. The
CEM II/B, respectively), a blastfurnace type III cement
(CEM III) and pozzolanic type IV cement (CEM IV). Natural camera was placed in a tripod and all the images were obtained
rounded sand rich in quartz, supplied by Eduardo Torroja using the automatic shot feature, so that no movement could
Institute for Construction Science (CEN-Standard Sand EN distort the images. The obtained pictures had a resolution of
196), was also used. Mortar bars were stored in their moulds 3200 × 2400 pixels with a colour depth of 24 bits, resulting in
in a climate chamber for 24 h (96% RH/20 ◦ C), and after image sizes of 4.57 MB. An example of the obtained images
demoulding they were cured under water at 20 ◦ C for 28 days. can be seen in figure 2.
After this period, the samples were cut into cubic pieces A black cardboard piece was used as background, see
(40 mm side) with a diamond saw. Two opposite faces of the figure 2. This was done to ensure that the sample could be
cubes were sealed with epoxy resin Sikaguard 62, to ensure easily identified and subtracted during the image-processing
that the degradation process was limited to a plane. stage. Such a procedure ensures a controlled background
Accelerated degradation was achieved by immersing histogram. The illumination of the sample is a decisive stage
the samples in concentrated ammonium nitrate solutions in this kind of application. It can increase the precision and
(460 g L−1 ) for 32 days, with a solution/sample volume ratio reliability of the method. Diffuse natural sunlight was used to
equal to 11. The samples were taken out of the solutions at illuminate the samples.

2
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

Figure 1. Scheme of the analysis procedure developed to obtain the degraded depths by automatic image analysis.

is made by analysing a circular white piece of plastic with

known dimensions (r = 15.13 ± 0.01 mm) displayed on each
image, as can be seen on the right side of figure 2. To
minimize errors, the sample and the reference piece must
have similar thickness to be in the same focal plane. First,
the reference piece is isolated from the image (figure 3(a)).
The best way of obtaining the dimensions of the piece is by
segmenting the image and calculating the area of the piece.
Image segmentation is made by automatic thresholding with
Otsu’s method [25], which chooses the value to minimize
the intraclass variance of the black and white pixels, to
obtain a binary image (figure 3(b)). Finally, holes in the
image are filled and the image is filtered using a circular
averaging two-dimensional (2D) filter with a radius of 5 pixels
(figure 3(c)). Finally, the area is computed and the conversion
Figure 2. Example of the images used by the designed algorithm.
factor is obtained. On average, the conversion factor is
35 pixels mm−1 . Therefore, the maximum resolution on
3.2. Image pre-processing measuring the degraded depths will be about 0.03 mm.
Subsequently, the sample alignment is checked to find
Images imported into MatlabTM are stored as an m × n × 3 whether the sample is aligned or not with the image borders.
data array that defines the red, green and blue colour intensity The final output of the procedure is a measurement of the
components for each individual pixel. As the main purpose degraded depth; therefore, it is quite necessary to ensure
of this work is to obtain degraded depth measurements, it is that the image is square. Since linear measurements are
necessary to calibrate the images, i.e. relate distances in the made, a non-square image may provide wrong degraded depth
image (pixels) to those in the sample (mm). Such operation values. This step is automatically made, by determining the

3
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

(a) (b) (c)

Figure 3. Procedure used for image calibration: (a) reference piece after being isolated, (b) binary reference piece and (c) final filled
reference piece.

(a) (b) (c)

(d) (e) (f )

Figure 4. Procedure for image alignment: (a) original image, (b) selected side of the sample, (c) filtered and segmented side, (d) border
detection by Sobel’s method, (e) longest line detected by the SHT procedure and (f ) final automatically rotated image.

angle formed by the sample borders with the image borders, [29, 30]. Through this procedure, it is possible to identify
combining both edge and line detection. First, the image lines in the edges of the sample. To ensure that the line
is filtered to enhance the border identification. A nonlinear identified is related to the sample border, only the longest one
2D median filter (NLMF) with a 75 × 75 neighbourhood is is considered. The SHT procedure also allows us to measure
used, as it is the most effective way of filtering when the the angle formed by that line with the coordinated axes, so that
goal is to reduce noise and preserve edges simultaneously the image can be rotated. An example of how the alignment
[26]. Each output pixel contains the median value from procedure works can be seen in figure 4.
the 75 × 75 neighbourhood around the corresponding pixel
in the input image. Thus, an average filtering window 3.3. Image processing
of approximately 2.1 mm is applied through the selected
part of the image. After filtering, a blurred image will be The main operations in the image-processing step are
obtained in which the information on the edges is preserved background subtraction and image segmentation. Both
and the unnecessary information (e.g. pores or aggregates) operations are based on the histogram analysis. If the
is eliminated by averaging with information concerning the histograms of each colour plane in an RGB image are analysed
cementitious matrix. separately, it can be seen that if the proper colour plane is
The image is then segmented by Otsu’s method, to chosen, it is possible to differentiate between the sample and
enhance the differences between the image border and the the background, as well as between the degraded zone and
background. Edges are identified by searching the local the sound zone. Since the greyscale depth value is 0, the
maxima in the image gradient, as described by Sobel’s method background is easily differentiated in each colour channel, see
[27]. In the final step, lines are identified by computing figures 5(b) and (d). However, the degraded and the sound
the Standard Hough Transform (SHT) of the binary image zone can only be distinguished by using the green channel

4
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

(b)

(a)

(c)

(d)

Figure 5. (a) Original image; colour channels and the corresponding histograms from an RGB image: (b) red channel, (c) green channel
and (d) blue channel.

(figure 5(c)). Thus, background subtraction and image accurate threshold levels, consequently obtaining poor image
segmentation can be done by selecting the proper colour segmentation and wrong measurements of the degraded depth.
channel. The first way to avoid such an error during segmentation
Primarily, the background must be subtracted from the is applying a nonlinear median filter to the image. As said
images. As mentioned above, either red or blue colour before, it is the most effective way of filtering an image when
channels allow good differentiation between the sample and the goal is to reduce noise and preserve edges simultaneously.
the background, so one of those channels is automatically Such noise in this case is caused by the aggregates, the pores
selected to continue the background subtraction step. The and all the superficial defects that may be present. The NLMF
greyscale image is then filtered to ensure better subsequent filter is applied with a 75 × 75 neighbourhood. As the filtering
thresholding. The NLMF filter with a 75 × 75 neighbourhood window is larger than the aggregate maximum size (2 mm), a
is used, as described in the pre-processing step. Then, the blurred image will be obtained with enhanced differentiation
background and the sample are merged by segmentation. The between the degraded zone and the sound zone, see figure 7.
threshold level is automatically computed by Otsu’s method To allow better processing and avoid errors associated with
and the greyscale image is converted to a binary image (pixel the corners of the sample, the image is cropped to obtain
values are 0 or 1), in which the background pixels have a four separated images (called si ) related to each side of the
value equal to 0. Finally, a morphological reconstruction [30] sample.
is performed in the binary image to obtain a uniform image. At this point, it is possible to compute an optimal threshold
This binary image is used to subtract the background from the level for the si image, so the degraded zone and the sound
original image. The whole process of background subtraction zone can be clearly distinguished. According to the kind
is shown in figure 6. of histogram displayed by the si image (figure 7(b)), the
The final step comprises the segmentation of the image in most suitable thresholding method is Rosin’s method [32, 33].
such a way that the pixels related to the degraded zone and the Rosin’s thresholding method is specifically intended to cope
sound zone can be distinguished. The final output shall be an with essentially unimodal distributions rather than the usual
image in which the pixels of both the non-degraded zone and bimodal or multimodal distributions. That is, the second peak
the background are equal to 0. The main problem arises when is either very small or is submerged within the main peak. It
in the image histogram there is no clear distinction between is worth noting that it is quite difficult for the thresholding
the zones of interest, as happens when trying to distinguish method to be suitable for all the images to be studied. In the
between the degraded zone and the sound zone (figure 5(c)). case under study, as the material is quite heterogeneous and
A bimodal distribution can be seen in the histogram, with the can stand for several microstructural characteristics (aggregate
peaks located approximately at greyscale values of 90 and type, porosity, mineral composition, superficial imperfections
200, but with no clear valley between both peaks. In such of the samples, non-uniform staining, etc), the images to be
cases, the automatic thresholding methods may not provide analysed can also change.

5
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

(b) (c)

(a)

(e) (d)

Figure 6. Procedure for background subtraction: (a) original image, (b) selection of the appropriate colour plane, (c) median filtering,
(d) segmentation and morphological reconstruction and (e) final background-subtracted image.

intensities. In addition, the degradation process may modify

the sample’s surface. All these features will be reflected in the
(a)
histogram, see figure 8.
In a first step, each si image is automatically segmented by
using Rosin’s thresholding method. When all the si images of
each sample have been processed, the segmentation process is
evaluated by the user. Those images in which the segmentation
has not been accurate will again start the segmentation process.
Four different thresholding methodologies will be considered
and evaluated: Otsu’s method, a fuzzy c-means thresholding
[33], Kapur’s method [35, 36] and the isodata algorithm
[37, 38]. At the end of the loop, a binary si image, bsi , is
obtained. This image is filtered with a 75 × 75 neighbourhood
(b)
NLMF filter and used to subtract the sound zone from the
si image. The whole segmentation process is shown in
figure 9.

3.4. Degraded depth measurements

As shown in figure 1, the final output of this methodology is
a measure of the degraded depth. To obtain such a measure,
an image in which the pixels of both the non-degraded zone
and the background are equal to 0 is required. This image, as
previously mentioned, is obtained from the image-processing
Figure 7. Grey level histograms of the images to be analysed:
(a) before applying the 75 × 75 NLMF filter and (b) after filtering. step. This kind of image will allow the degraded depths
from pixel-value profiles across the x-axis of the images to
be obtained. An example of such an image is shown in
Thus, the differentiation between the degraded zone and figure 10(a). If each n column (the x-axis) of the final
the sound zone may not be as clear as shown in figure 7(b) segmented si image is plotted versus depth, as shown in
even after filtering. An example of such a situation is figure 10(b), it is possible to obtain colour intensity profiles
the effect of different cement types on the images to be along the thickness of the sample.
processed. The stain used to identify the degraded zone— Since the pixels of both the non-degraded zone and the
phenolphthalein—depends on the pH of the pore solution background are equal to 0, the colour intensity profile of the
and other microstructural features of the material. Different pixels along each n column of si will be a step function.
cement types may display different colour shades and Thus, it is possible to obtain a degraded depth value for

6
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

(a)

(b)

(c)

(d)

Figure 8. Examples of different stained mortar samples and their histograms: (a) CEM I sample—w/b 0.6—32 days of degradation,
(b) CEM II/B sample—w/b 0.4—32 days of degradation, (c) CEM IV sample—w/b 0.6—11 days of degradation and (d) CEM IV
sample—w/b 0.4—21 days of degradation.

(a) (b)
(c)

(d)

(e)

Figure 9. Procedure for the image segmentation: (a) obtaining of the si image by cropping, (b) selection of the colour channel and filtering,
(c) thresholding by Rosin’s method and filtering, (d) segmentation of the si image and (e) evaluation of the alternative thresholding methods
and choice of the most adequate one (in this case, Otsu’s method).

each n column of si , just by determining the full width each si image êD and its standard deviation σ d can be easily
at half maximum (FWHM) of each colour intensity profile. obtained. It is worth noting that each value of êD will be the
FWHM is a statistical parameter commonly used to describe a mean value of the n degraded depth values, where n is the
measurement of the width of an object in a picture, when that dimension in pixels through the x-axis of the si image. In
object does not have sharp edges. It is given by the distance the example shown in figure 10, the image has an n value
between points on the curve where the function reaches half of 945 pixels. The statistical meaning of the degraded depth
of its maximum value. The mean degraded depth value for measurements made by using this method is thus evident.

7
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

(a) (b)

Figure 10. Final step of the procedure: parameter extraction; (a) segmented si image and (b) colour intensity profile along the thickness of
the sample.

4. Results and discussion between the degraded and non-degraded zones. In all cases,
the combination of low staining intensity and dark areas of
The algorithm designed has been used to analyse and process the degraded zone (caused by its high porosity or resulting
the photographs of the materials described in section 2. from the cutting process) does not allow us to obtain accurate
The histograms of the samples shown in figure 8 seemed segmentations using a fully automatic method.
to indicate poor results at the end of the segmentation The degraded depths obtained with the designed algorithm
process. The automatic segmentation process was applied to are shown in table 2. The maximum relative standard deviation
those samples, and the so-obtained results are presented in of the degraded depth measurements using the automatic
figure 12. The errors generated during the segmentation method presented here is about 6%. This value is associated
process were confirmed in some cases. In practice, it is with the repeatability of the measurements. If microstructural
very difficult to develop an automatic method that can take variations and differences between samples are considered, the
into account all the processes and characteristics associated algorithm provides very good results. The precision (about
with this kind of material. We found that porosity, superficial 0.03 mm) and statistical representation of the measurements
imperfections of the samples and non-uniform staining may made with this automatic procedure was not attained by other
induce such errors during automatic processing of the images. methods used to date.
Supervision is thus needed to identify the samples that may go The measurements of the degraded depths made with
through a semi-automatic processing step. the developed procedure shall be contrasted, to determine
Figure 11(a) shows a sample with w/b 0.6 and 32 days whether the segmentation was successful or not. This is
of degradation of CEM I. Analysis of the histogram conducted by comparing the automatic measurements with
corresponding to the filtered greyscale image (figure 8(a)) the visual measurements of the degraded depth. Manual
indicates that there is no clear difference between the peak measurements are the most used procedure to determine
corresponding to the degraded zone (centred around 145) and degraded or carbonated depths [19, 20]. These measurements
the one corresponding to the sound zone (centred around are usually made by directly measuring the thickness of the
94). Thus, the thresholding process will not give an interesting zone with a calliper. In this case, using the same
accurate segmentation of the image, mainly due to porosity images as those used by the algorithm, the degraded depths
of the sample, which introduces dark zones in the image. were determined by manually measuring the distance in pixels
Figure 11(b) shows a sample with w/b 0.4 and 32 days of between the border of the sample and the beginning of the
degradation of CEM II/B. In this case, the situation is more stained zone. Five measurements were made along each side
complex since in some parts of the image it is quite difficult to of the samples, avoiding the corners of the sample and trying to
differentiate between the degraded zone and the sound zone. In have a homogeneous distribution of measurements. There are
this case it is not possible to obtain good image segmentations very good correlations between the automatically measured
using the automatic procedure, mainly due to the low contrast and the visually measured values (figure 12). The correlation

8
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

(a)

(b)

(c)

(d)

Figure 11. Results of the segmentation process in the samples shown in figure 8: (a) CEM I sample—w/b 0.6—32 days of degradation,
(b) CEM II/B sample—w/b 0.4—32 days of degradation, (c) CEM IV sample—w/b 0.6—11 days of degradation and (d) CEM IV
sample—w/b 0.4—21 days of degradation.

Table 2. Degraded depths obtained with the algorithm and standard deviations of the measurements.
Degraded depths (mm)
w/b ratio Time (days) CEM I CEM II/A CEM II/B CEM III CEM IV

0.4 4 3.22 0.16 2.53 0.12 2.29 0.10 2.89 0.09 2.72 0.04
11 5.73 0.14 4.18 0.01 3.74 0.13 5.01 0.15 – –
21 6.19 0.19 5.47 0.10 6.96 0.04 6.78 0.20 5.79 0.11
32 8.31 0.29 7.51 0.08 9.95 0.19 8.97 0.07 7.97 0.18
0.5 4 3.26 0.11 2.73 0.05 2.98 0.04 3.26 0.07 3.24 0.19
11 5.81 0.21 4.36 0.09 4.33 0.04 5.27 0.14 5.13 0.19
21 8.41 0.10 5.75 0.14 7.46 0.19 6.66 0.06 7.14 0.30
32 11.68 0.27 7.60 0.08 10.14 0.07 8.45 0.20 8.72 0.15
0.6 4 3.88 0.10 3.24 0.19 3.27 0.08 3.40 0.16 3.81 0.06
11 4.81 0.19 4.37 0.08 5.94 0.05 6.28 0.22 4.53 0.28
21 8.33 0.13 7.24 0.09 7.58 0.10 7.35 0.10 8.13 0.28
32 14.33 0.10 8.41 0.09 9.37 0.15 10.39 0.14 11.43 0.26

factor expressed as the adjusted-R2 of the linear fit is 0.96. residuals of the automatically measured degraded depths are
This shows the merit of the automatic procedure designed to plotted versus each measured value, as shown in figure 13. It
measure the degraded depth. The mean error, expressed as the can be seen that there are no anomalous trends, and just a few
difference between the automatically measured values and the studentized residuals exhibit an absolute value greater than 1.
visually measured values, is about 6%. The samples matching to those outlier measurements will be
analysed separately.
The analysis of the residuals can provide very useful Figure 14 shows the samples whose studentized residual
information regarding which samples provide the highest values were greater than 1. Some of the samples were
errors, and thus identify the limitations of the algorithm already shown in figure 11. The explanation for the outlier
designed. With this aim the analysis of studentized residuals measurements is the low contrast between the degraded and
will be used to identify which measures are statistically distant non-degraded zones of those samples. As mentioned above,
from the rest of the data (outlier values). The studentized the combination of low staining intensity and dark areas of

9
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

Figure 15. Example of a concrete sample subjected to a chloride

migration test and sprayed with a 0.1 N AgNO3 aqueous solution.

segmentations using a fully automatic method. At this point

it is difficult to stipulate how to solve these situations. The
Figure 12. Correlations between the automatically measured and material’s heterogeneity does not allow us, most of the times,
the visually measured degraded depths, for all the studied samples. to systematize a procedure for obtaining accurate staining and
avoid superficial imperfections. However, a research group
has recently published some works describing an experience-
based reasoning approach [38, 39], which may be applicable
to solve this problem. It uses lineal combinations of each
colour plane to obtain an enhanced image; the coefficients for
the lineal combination are obtained by genetic algorithms to
obtain the most appropriate segmentation.
It is worth noting that the algorithm designed can also
serve to measure the degraded depths in samples subjected
to other degradation processes, such as chloride ingress.
Figure 15 shows a concrete sample that was subjected to a
chloride migration test. At the end of the experiment, the
sample was cut in half and sprayed with a 0.1 N AgNO3
aqueous solution. The chloride-affected zone appears white,
whereas the chloride-free zone is red-brown, enabling the
detection of the borderline between the two zones.
The automatic segmentation process was applied to
this sample, and the so-obtained results are presented in
Figure 13. Studentized residuals versus automatically measured figure 16. The result of the segmentation is quite satisfactory;
degraded depths, for all the studied samples.
minor errors were obtained due to the characteristics of the
aggregates. This problem may also appear when applying
the degraded zone (caused by its high porosity or resulting the designed algorithm to measure the degraded depths in
from the cutting process) does not allow us to obtain accurate concrete samples. Aggregates may introduce some difficulties

(a) (b) (c) (d)

(e) (f) (g) (h)

Figure 14. Samples matching the outlier values shown in figure 13.

10
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

(a) (b)

Figure 16. Applications of the algorithm designed to measure the depth of chloride penetration.

in the automatic segmentation process, because they usually Ministry of Public Works, through research project C01/2007,
do not get coloured by the stain and they can have colour is gratefully acknowledged. Dr M Molero was supported by
intensities similar to the background or the non-degraded the department of education of the Community of Madrid
zone. We are actually working to implement a final code and the European Social Fund. Dr S Aparicio is supported
which allows the degraded depth measurement in cementitious by the post-doctoral JAE-DOC program of the Spanish
materials with different shapes, high aggregate content and National Research Council (CSIC). The authors also wish to
subjected to different degradation processes. acknowledge the contribution of Dr David Martı́n Gómez and
Dr Xavier P Burgos Artizzu for their help and advice on image-
5. Summary processing techniques.

A novel methodology for measuring the degraded depths in References

cementitious materials by combining a staining procedure
and an image-processing algorithm has been presented. It [1] Poole A B, McLachlan A and Ellis D J 1988 A simple staining
is possible to reveal the state of a pH-dependent degradation technique for the identification of alkali-silica gel in
concrete and aggregate Cement Concr. Res. 18 116–20
process in cementitious materials using phenolphthalein. [2] Redon C, Chermant L, Chermant J and Coster M 1999
Degraded materials are cut, stained and photographed. It Automatic image analysis and morphology of fibre
is possible to obtain the degraded depths through further reinforced concrete Cement Concr. Compos. 21 403–12
image analysis. Special attention is devoted to the sample’s [3] Jakobsen U H and Brown D R 2006 Reproducibility of w/c
ratio determination from fluorescent impregnated thin
illumination to obtain good images.
sections Cement Concr. Res. 36 1567–73
The designed algorithm combines several operations to [4] Ammouche A, Breysse D, Hornain H, Didry O and
calibrate and to align the images, to eliminate the background Marchand J 2000 A new image analysis technique for the
and to carry out further thresholding to segment the image, quantitative assessment of microcracks in cement-based
in an automated way. The images used have a resolution of materials Cement Concr. Res. 30 25–35
[5] UNE-EN 14630 2007 Products and systems for the protection
3200 × 2400, allowing a measurement precision of about and repair of concrete structures. Test methods.
0.03 mm/pixel. Using the developed procedure, it is Determination of carbonation depth in hardened concrete
possible to measure the degraded depths for a set of by the phenolphthalein method (Madrid: AENOR)
different cementitious materials, corresponding to different [6] Nguyen V H, Colina H, Torrenti J M, Boulay C and Nedjar B
cement types, w/b ratios and degradation times. In all the 2007 Chemo-mechanical coupling behaviour of leached
concrete: part I. Experimental results Nucl. Eng. Des.
studied cases, the maximum relative standard deviation of the 237 2083–9
measurements was about 6%. The results were compared [7] RILEM 1988 CPC-18 measurement of hardened concrete
with visual measurements of the degraded depth, to contrast carbonation depth Mater. Struct. 21 453–55
the results. Very good correlations were found in all the [8] Chang J J, Yeih W, Huang R and Chen C T 2004 Suitability of
cases. The mean error, expressed as the difference between several current used concrete durability indices on
evaluating the corrosion hazard for carbonated concrete
the automatically measured values and the visually measured Mater. Chem. Phys. 84 71–8
values, is about 6%. The designed algorithm can also [9] Lawrence R M H, Mays T J, Walker P and D’Ayala D 2006
be applied to measure the degraded depths of cementitious Determination of carbonation profiles in non-hydraulic lime
materials subjected to other degradation processes, such as mortars using thermogravimetric analysis Thermochimica
chloride ingress. Acta 444 179–89
[10] Jones M, Dhir R, Newlands M and Abbas A 2000 A study of
the CEN test method for measurement of the carbonation
Acknowledgments depth of hardened concrete Mater. Struct. 33 135–42
[11] Houst Y F and Wittmann F H 2002 Depth profiles of
carbonates formed during natural carbonation Cement
Support from the Spanish Science and Innovation Ministry, Concr. Res. 32 1923–30
through research projects BIA2006-15188-C03-01, BIA2009- [12] Perlot C, Verdier J and Carcassès M 2006 Influence of cement
14395-C04-01 and BIA2010-16510, and from the Spanish type on transport properties and chemical degradation:

11
Meas. Sci. Technol. 21 (2010) 055103 I Segura et al

application to nuclear waste storage Mater. Struct. Engineering Department: Construction. School of Civil
39 511–23 Engineering of Madrid
[13] Petrov I and Schlegel E 1994 Application of automatic image [25] Otsu N 1979 A threshold selection method from gray-level
analysis for the investigation of autoclaved aerated concrete histograms IEEE Trans. Syst. Man Cybern. 9 62–6
structure Cement Concr. Res. 24 830–40 [26] Lim J S 1990 Two-Dimensional Signal and Image Processing
[14] Chermant J 2001 Why automatic image analysis? An (Englewood Cliffs, NJ: Prentice-Hall)
introduction to this issue Cement Concr. Compos. [27] Parker J R 1997 Algorithms for Image Processing and
23 127–31 Computer Vision (New York: Wiley)
[15] Chermant J, Chermant L, Coster M, Dequiedt A and Redon C [28] Hough P V C 1962 Method and means for recognizing
2001 Some fields of applications of automatic image complex patterns US Patent 3,069,654
analysis in civil engineering Cement Concr. Compos. [29] Duda R O and Hart P E 1972 Use of the Hough transformation
23 157–69 to detect lines and curves in pictures Commun. ACM
[16] Dequiedt A, Coster M, Chermant L and Chermant J 2001 15 11–5 Available at http://doi.acm.org/10.1145/
Distances between air-voids in concrete by automatic 361237.361242
methods Cement Concr. Compos. 23 247–54 [30] Soille P 1999 Morphological Image Analysis: Principles and
[17] Soroushian P, Elzafraney M and Nossoni A 2003 Specimen Applications (Berlin: Springer)
preparation and image processing and analysis techniques [31] Rosin P L 2001 Unimodal thresholding Pattern Recognit.
for automated quantification of concrete microcracks and 34 2083–96
voids Cement Concr. Res. 33 1949–62 [32] Conaire C Ó 2007 Code implementation of Rosin’s
[18] Basheer L, Basheer P A M and Long A E 2005 Influence of thresholding method Available at http://elm.eeng.dcu.ie/
coarse aggregate on the permeation, durability and the ∼oconaire/source/
microstructure characteristics of ordinary Portland cement [33] Xiong G 2005 Fuzzy c-means thresholding Available at
concrete Constr. Build. Mater. 19 682–90 http://www.mathworks.es/matlabcentral/fileexchange/8351
[19] Chang C and Chen J 2006 The experimental investigation of [34] Kapur J N, Sahoo P K and Wong A K C 1985 A new method
concrete carbonation depth Cement Concr. Res. for gray-level picture thresholding using the entropy of the
36 1760–7 histogram Comput. Vis. Graph. Image Process. 29 273–85
[20] Segura I, Anaya J J, Hernández M G, Macphee D E, [35] Conaire C Ó 2007 Code implementation of Kapur’s
Moragues A and Sánchez T 2007 Microstructural thresholding method Available at http://elm.eeng.dcu.ie/
characterization of aged mortar by destructive and ∼oconaire/source/
non-destructive testing 12th Int. Congress on the Chemistry [36] Ridler T W and Calvard S 1978 Picture thresholding using an
of Cement (Montreal, Canada) iterative selection method IEEE Trans. Syst. Man Cybern.
[21] Segura I, Moragues A, Macphee D E, Anaya J J and Molero M 8 630–2
2009 Study of the decalcification process in mortars [37] Ramachandram D 2003 Automatic thresholding Matlab
degraded by NH4 NO3 by using ultrasonic techniques Central Available at http://www.mathworks.com/
Mater. Construc. 59 17–36 matlabcentral/fileexchange/3195
[22] UNE-EN 196-1 1996 Methods of Testing Cement. [38] Burgos-Artizzu X P, Ribeiro A, Tellaeche A, Pajares G
Determination of Strength (Madrid: AENOR) and Fernández-Quintanilla C 2009 Improving weed
[23] Heukamp F H, Ulm F and Germaine J T 2001 Mechanical pressure assessment using digital images from an
properties of calcium-leached cement pastes. Triaxial stress experience-based reasoning approach Comput. Electron.
states and the influence of the pore pressures Cement Concr. Agric. 65 176–85
Res. 31 767–74 [39] Burgos-Artizzu X P, Ribeiro A, Tellaeche A, Pajares G
[24] Segura I 2008 Characterization of the decalcification process and Fernández-Quintanilla C 2010 Analysis of natural
in mortars by means of destructive and non-destructive images processing for the extraction of agricultural
testing PhD Thesis Polytechnic University of Madrid, Civil elements Image Vis. Comput. 28 138–49

12