AN UPDATE ON APPLICATION OF A FLAT-BED SCANNER FOR
PERFORMING ASTM C 457
J. Carlson, L. Sutter, K. Peterson, T. Van Dam
Michigan Tech Transportation Institute
Michigan Tech University
Houghton, MI - USA
ABSTRACT
Determination of the air-void system parameters of hardened concrete is generally
performed using an optical microscope and procedures outlined in ASTM C 457. This
process is tedious and the results dependant upon the operator performing the analysis.
Recently, an approach to performing this analysis has been developed using a highresolution flat-bed scanner. The developed approach provides a complete analysis
including aggregate and paste fractions based upon image analysis techniques. This
paper presents recent developments at streamlining this analysis process to provide
fundamental analysis of the air-void system parameters assuming aggregate and paste
contents from mix design data. Additionally, the results obtained from the highresolution scanner are compared with results from the same specimens obtained from a
conventional office flat-bed scanner.
INTRODUCTION
In the past few years, a great deal of research has been performed to develop an
automated system to perform ASTM C 457, “Standard Test Method for Microscopical
Determination of Parameters of the Air-Void System in Hardened Concrete.” The
various methods are similar in the respect that an image of the concrete specimen is
collected, analyzed and summarized with little aid of a human operator [2, 3, 4, 6]. The
differences between the methods occur in the manner in which that image is collected.
This paper presents an advancement to an automated method based upon image
collection with a high resolution flat-bed scanner.
The preceding method utilized a high resolution flat-bed scanner with 8 x 3 micron (3175
x 8000 dpi) capability. Three images of a polished concrete specimen were collected by
the scanner. The first image collected was the unaltered, polished surface of the concrete.
Then, the specimen’s surface was coated with a phenolphthalein solution which indicated
pink upon contact with the hydrated cement paste. This surface was scanned as the
second source image. Finally, the third image was obtained by the scanner after the
surface was contrast enhanced by filling the voids white and coloring the remainder
black. The three images were combined and processed to determine the standard air void
system characteristics as described in Peterson et al. [4].
The method presented in this paper utilizes only the final of the three surface images
briefly described above. The ‘black and white’ contrast enhanced image was used in
conjunction with knowledge of either the mix design or batch weights to perform the
calculations set forth in ASTM C 457. This analysis was accomplished in about 15-20
minutes using a computer script program written for widely available and relatively
inexpensive software packages on a standard personal computer. In addition, a
comparison was made between the quality of images obtained from the high resolution
flat-bed scanner and a conventional office flat-bed scanner with an advertised optical
resolution capability of 4800 x 4800 dpi (~5.3 x 5.3 micron).
EXPERIMENTAL
Initial sample preparation
This study was conducted upon eighteen lab produced specimens of varying mixture
proportions, air entraining admixtures and target air contents. Four by eight inch
standard test cylinders were prepared from each of the eighteen mix designs. A slab of
approximately 100 mm x 200 mm x 20 mm (4” x 8” x ¾”) was cut perpendicular to the
finished surface of the cylinders using a kerosene cooled diamond blade saw. Then, one
face was polished on a lapping wheel. A series of diamond abrasive magnetic backed
platens were used with water as the coolant/cleansing medium. The slab was polished
using successively smaller grit abrasive down to, and including, 200 grit. Between each
grit size, short blasts of compressed air along with a water rinse were applied to ensure
polishing residuals did not accumulate in the voids. Next, to stabilize the voids with the
goal of obtaining the sharpest void edges possible, a solution of 5:1 by volume acetone
and fingernail hardener was applied to the partially polished surface [2]. This solution
was brushed on to the dried surface in two coats applied perpendicular to each other.
Once the fingernail hardener had set, lap wheel polishing continued with 600 grit
adhesive backed silicon carbide paper. Finally, the polished surface was bathed in
acetone for one minute to remove the remaining fingernail hardener. To complete the
initial surface preparation, twelve 6 mm x 3 mm x 200 micron thick reflective stickers
were applied to the perimeter of the polished surface. These stickers were used to protect
the glass surface of the flat-bed scanners and also as location reference points. The
reference points allowed for easy alignment and comparison of data collected during the
standard motorized stage point count, the high-resolution flat-bed scanner image analysis
and the conventional office flat-bed scanner image analysis.
Motorized stage manual point count
Reference values for the air void systems of each of the specimens needed to be known to
determine the accuracy of the flat-bed scanner based analyses. In this study, the methods
described in ASTM C 457 Procedure B – Modified Point Count Method were used as the
standard to which the scanner based analyses were compared [1]. A stereo microscope,
CCD and motorized stage were used to perform the manual modified point count. These
components were connected to a computer and operated with the assistance of a macro.
The macro was written for NIH Image which is an image analysis program developed by
the United States National Institutes of Health available for download on the web
(www.scioncorp.com) [5]. Along with performing the modified point count, the macro
allowed for the x,y coordinates and corresponding phase (hydrated cement paste,
aggregate or air void) of each of the 1,388 stops in the point count to be recorded. These
coordinates and observed phases were used to perform an accuracy assessment between
the images collected by the scanners and the data recorded by the operator performing the
motorized stage point count.
Surface contrast enhancement
The specimens were contrast enhanced, or colored black and white, once the modified
point count was complete. The steps and materials used to obtain a contrast enhanced
image vary for the different laboratories that perform such analyses, but the desired
results are the same [2, 3, 4, 6]. The goal of the contrast enhancement was to produce a
surface where the voids in the concrete are filled with a white medium while the
remaining phases (cement paste and aggregate) were colored black. Ideally, this contrast
allows the image capturing device used to clearly distinguish between the portion of the
sample that are voids and the non-voids. The steps used in this study were as follows:
1. A 16mm x 8 mm tip black permanent marker was used to apply the initial coat of
ink to the surface. The ink was applied in overlapping parallel lines across the
sample being careful to avoid marking the hologram stickers.
2. The same marker was used to apply a second coat of ink. This coat was applied
in the same overlapping fashion but perpendicular to the initial coat.
3. The ink on the sample was allowed to dry long enough to ensure it would not be
removed or smudged in the following steps.
4. The surface was covered with a thin layer (~ 2mm) of 2 micron wollastonite
powder.
5. The flat portion of a glass slide was used to press the powder into the voids in the
sample.
6. A razor blade edge was gently passed across the surface to remove the bulk of the
excess wollastonite powder.
7. Finally, a very slightly oiled index finger was gently passed across the surface to
remove the remaining residual wollastonite powder.
8. The contrast enhancement quality was checked using a stereo microscope
ensuring the voids were sufficiently permeated and all residual wollastonite was
removed.
Image collection via flat-bed scanner
To perform the automated C 457 analysis, a TIFF image of the specimen was collected at
a pixel resolution of 8 x 8 microns. This study utilized a high resolution flat-bed scanner
with a proven capability of obtaining such a resolution along with a previously untested
conventional office flat-bed scanner claiming to be capable of the same resolution. The
high resolution flat-bed scanner used in the study was a CreoScitex EverSmart Pro II
featuring Scitex Stitch scanning technology with 3 x 8 micron (3,175 x 8,000 dpi)
maximum optical resolution capability. The conventional office flat-bed scanner used
was a Hewlett-Packard 8200 Scanjet with 5.3 x 5.3 micron (4,800 x 4,800 dpi) maximum
optical resolution capability. Images were obtained from the 18 specimens by each of the
scanners in grayscale (256 shade) mode at a resolution of 8 by 8 microns. The high
resolution scanner took approximately 30 minutes to obtain the image while the
conventional office scanner took 10 minutes. The difference was mainly attributed to the
high resolution scanner making two passes over the sample while the conventional office
scanner made a single pass in collecting the image. Figure 1 displays a comparison of the
image quality attained by the high-resolution and office scanner respectively.
Figure 1: Image from High-Resolution Scanner (left) and image from Conventional Office Scanner (right)
Scripting an automated point count
A computer script program was developed to calculate the standard air void system
parameters from ASTM C 457 using the grayscale image. This script was written in the
Visual Basic Script language which takes advantage of the ability to command multiple
Microsoft Windows applications within a single script program. This script is
programmed to utilize Adobe’s Photoshop CS, Microsoft’s Excel and Microsoft’s Word
in performing a modified point count upon the image and outputting a report file. The
steps that the script performs, and a discussion of each, are as follows:
1. The operator enters identifying information about the specimen.
This information includes the specimen’s identification, source, technician’s name and
other miscellaneous information to be presented on the output report.
2. The operator enters either the batch weights or volumetric percent paste in the
concrete specimen.
This is an important step in being able to perform this analysis. In order to perform the
ASTM C 457 calculations, the volume of paste in the sample must be known. In a
standard point count, the operator simply keeps track of what percentage of stops fall
upon the paste. In a linear traverse, the length of the traverse line passing over the paste
is recorded. These methods are not viable when analyzing the contrast enhanced
specimen since the black portion of the specimen consists of both the aggregate and paste
phases of the concrete. Therefore, this script uses the theoretical amount of paste that
should be in the sample based on the batch weights of the concrete. The specific
gravities of the materials must be known to perform the volumetric calculations, and
should be obtained from the material supplier. The calculated paste volume in most cases
will be sufficiently close to the actual value. Even if slight variations in the volume of
the paste do occur, it has been shown that the effect upon the air void system parameters
is minimal [3].
3. The grayscale image is opened in Photoshop and the desired portion of the
specimen to be analyzed is selected by cropping out the undesired portions.
This step allows the operator to select the desired area and avoid the portions of the
image that may be undesirable for analysis such as the reflective stickers.
4. The operator selects a threshold value for the grayscale image.
The threshold step is by far the most important step in performing C 457 on a contrast
enhanced image. The scanned grayscale image of the sample consists of many pixels,
each with its own color falling in the 256 grayscale range between pure black and pure
white.
To compare the air void system of specimens against each other, a consistent threshold
value needs to be used. It is believed that this threshold value needs to be determined on
a lab to lab basis, since each lab will have different environments (particularly lighting)
in which their flat bed scanners will be operating. In addition, varying scanners and/or
surface preparation techniques could alter the ideal threshold value for the image.
The importance of selecting a threshold for an ASTM C 457 analysis is that it becomes
the dividing line between what is and what is not air. Figure 2 depicts a series of images
adjusted with increasing threshold values. The amount of ‘air’, or white pixels, in the
image decreases as the threshold value is increased.
Figure 2: Example Threshold Value Comparison
A simple method to determine the best threshold value to use in performing the ASTM C
457 analysis was developed and is presented later on in this report.
5. The operator is given the opportunity to mask any internal voids within aggregate
particles using the image software tools.
The aggregates used in portland cement concrete mixes are variable. Their hardness,
chemistry and angularity are all characteristics that are important to concrete, but the
amount of internal voids is very important in this type of analysis. A plane section of
concrete will intersect a great number of individual aggregate particles. Depending on
the source of the aggregate, a number of these particles will have interior macroscopic
voids. The operator of a standard point count or linear traverse can simply use judgment
to ignore the voids within the aggregate. However, in the surface preparation of the
sample for automated analysis, these voids will become filled as do the entrained air
voids. Thus, the scanner will report such pixels in an image as ‘white’ pixels. The
operator is prompted to use the Photoshop tools to mask these voids to avoid allowing the
script to count these pixels as ‘air’. Figure 3 displays an example of an aggregate particle
before and after the masking operation. The masking step can, in cases of concretes
composed entirely of porous aggregates, take up to an hour to perform. For this study, in
which a partially crushed natural gravel was used, the masking operation took
approximately five to ten minutes per image to perform. An alternate method of
effectively masking the aggregate voids is by marking the voids with a black marker prior
to scanning.
Figure 3: Masking of Internal Air Voids in Aggregate Particle
6. The specific regions of the image to be analyzed per ASTM C 457 are selected.
Based upon the size of the aggregate used in the mixture, ASTM C 457 requires both a
certain number of points to be counted and length of line to be traversed in performing
the modified point count. The script selects and removes the required strips of pixels
from the full image and creates a new composite image of these points as shown in
Figure 4. The composite image contains fewer pixels than the scanned image and is thus
analyzed more quickly while still meeting ASTM requirements.
Scanned Image
Masked Image
Strips Removed
Composite Image Size
Figure 4: Steps in Creation of Composite Image by Script
7. The selected threshold value is applied to the composite image to convert the
image to binary format.
The pixels in the image are converted from the 256 grayscale values collected by the flatbed scanner to binary values per the selected threshold value. Figure 5 shows an example
of sixteen pixels with varying pixel intensities being converted to a binary image.
Figure 5: Example Threshold Conversion from Scanned to Binary Pixel Intensities
8. The binary image is analyzed per ASTM C 457 using a spreadsheet as the
analysis tool.
As in Figure 6, the binary image is exported from the imaging software to a spreadsheet
in such a fashion that each of the cells in the spreadsheet corresponded with a single pixel
from the image.
Figure 6: Example Threshold Conversion from Scanned to Binary Pixel Intensities
The modified point count is then performed upon the
cells to determine the parameters of the air void
system. Each cell is considered to be a stop in the
point count portion of the test and is counted as
either air (for cells = 0) or not air (for cells = 255).
Dividing the total count of cells classified as air by
the total number of cells gives the volume percent air
in the sample. Each column in the spreadsheet is
considered for the determination of the number of
voids intersected for the modified point count. The
quantity of the voids intercepted was obtained along
with the length of each void, or chord, intercepted.
Figure 7 depicts an example of this portion of the
analysis. If a count was being performed on these
pixels alone, the sample would have an air content of
56.25% (9 out of the 16 pixels are white). This
example also has three intercepted voids with
lengths of 8, 16 and 48 microns.
Figure 7: Void Intercept Analysis
Once the data is collected, the calculations for void frequency, paste to air ratio, average
chord length, specific surface and spacing factor are performed. In addition, collecting
the lengths of all the intersected air voids allows for easy plotting of the chord length
distribution.
9. Results are presented in report format with the aid of word processing software.
The identifying characteristics of the sample, calculated air void system parameters and
chord length distribution graphs are presented in a report format.
Threshold value selection
An ideal threshold value was determined for each of the two scanners tested in this study.
The ideal threshold value for each scanner was found by running the automated script on
each of the 18 samples over a range of threshold settings. The resulting air void system
parameters for each threshold value were compared to the parameters calculated during
the motorized stage analysis. The threshold value with the best correlation in respect to
volume of air and spacing factor for each respective scanner was chosen as the ideal
threshold value for that particular scanner. Figure 8 depicts linear regressions of the
parameters obtained from the motorized stage analyses to those calculated by the
automated script analyses. Determined ideal threshold values were 180 and 160 for the
high-resolution and office flat-bed scanners respectively.
Volume % Air
Powers Spacing Factor
2
2
R = 0.7620
R = 0.7308
0.800
Office Flat-Bed Scanner
Office Flat-Bed Scanner
16.0
12.0
8.0
4.0
0.0
0.0
4.0
8.0
12.0
0.600
0.400
0.200
0.000
0.000
16.0
Motorized Stage (Manual Point Count)
0.200
0.400
0.800
Motorized Stage (Manual Point Count)
Volume % Air
Powers Spacing Factor
2
R = 0.7716
2
R = 0.7221
0.800
High Resolution Scanner
16.0
High Resolution Scanner
0.600
12.0
8.0
4.0
0.600
0.400
0.200
0.000
0.0
0.0
4.0
8.0
12.0
16.0
Motorized Stage (Manual Point Count)
0.000
0.200
0.400
0.600
0.800
Motorized Stage (Manual Point Count)
Figure 8: Air Void System Parameter Regression
Scatter plots to examine the correlation existing between the parameters calculated from
the two scanners were prepared. Figures 9 and 10 present the plots of the parameters
resulting from analysis of images from each of the scanners.
Scanner to Scanner Volume % Air
12.0
2
R = 0.9705
10.0
8.0
6.0
4.0
2.0
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
Conventional Office Flat-Bed Scanner
Figure 9: Correlation between Scanners – Volume % Air
Scanner to Scanner Air Void Spacing Factor
0.250
High Resolution Flat-Bed Scanner
High Resolution Flat-Bed Scanner
14.0
0.200
2
R = 0.9290
0.150
0.100
0.050
0.000
0.000
0.050
0.100
0.150
0.200
0.250
0.300
Conventional Office Flat-Bed Scanner
Figure 10: Correlation between Scanners – Air Void Spacing Factor
0.350
Accuracy assessment
An accuracy assessment between the data collected during the motorized stage point
count and data from the flat-bed scanners was conducted [7]. This assessment was used
to compare the quality of the results obtained from the flat-bed scanners to the manual
point count and also to compare the scanners to each other. As previously mentioned, the
x,y coordinates, and corresponding phases, for each of the 1,388 stops were recorded by
the macro used during the manual point count. Additionally, coordinates of distinctive
imperfections in the reflective stickers were noted. These reflective stickers were used to
correlate the coordinate system of the motorized stage to that of the scanned image of the
concrete sample. This alignment allowed the phase data collected on the motorized stage
to be overlaid upon the scanned image.
The manual point counts consisted of 1,388 points each in order to meet the ASTM C
457 requirement for minimum sample area analyzed. Pixels classified as air by the flatbed scanners were directly compared to points classified as air in the manual point count.
Similar comparisons were made for non-air pixels. The total number of pixels from the
18 samples correctly and incorrectly identified by each scanner was totaled. The results
of the pixel comparisons from the 18 high resolution flat-bed scanner images are
presented in Tables 1 and 2.
Table 1: Summary of High Resolution Scanner Accuracy Assessment
Motorized Stage (Manual Point Count)
Air Stops
Non-air Stops
High Resolution
Flat-Bed
1,262
2,109
Air Pixels
Scanner
320
21,293
Non-air Pixels
22,555/24,984 (90.28%)
Agreement between all pixels
Table 2: Summary of Office Flat-Bed Scanner Accuracy Assessment
Motorized Stage (Manual Point Count)
Air Stops
Non-air Stops
Office Flat-Bed
1,289
2,300
Air Pixels
Scanner
293
21,102
Non-air Pixels
22,391/24,984 (89.62%)
Agreement between all pixels
Discussion
A Visual Basic Script program was successfully developed and employed upon the
contrast enhanced concrete surfaces to perform ASTM C 457. The most crucial factor in
the accuracy of the calculated air void system parameters was the quality of the image
collected by the flat-bed scanner. Methods described in this report were used to compare
the capabilities of a high resolution flat-bed scanner to those of a conventional office flatbed scanner. Direct comparisons between the air void system parameters obtained from a
motorized stage manual point count and those from the automated script program were
made. Linear regression illustrated consistent correlations when comparing the results
from either scanner to the manual point count. Accuracy assessments comparing
individual pixels from the motorized stage manual point count to equivalent pixels in the
scanned images confirmed the correlations. The results show potential for an ASTM C
457 test procedure to be implemented using a flat-bed scanner based technique.
Although correlations exist between the scanner based method and the manual method, it
is clear from Figure 8 that the scanner based method consistently underestimates the
spacing factor when the manually determined values exceed 0.2 mm. The cause of the
disparity is the flat-bed scanner method’s tendency to calculate a higher void frequency
(voids intersected per unit length) than the manual point count, as illustrated by the
accuracy assessment. During the manual point counts performed on the 18 concrete
samples, a total of 1,582 points were identified as air, and a total of 23,402 points were
identified as non-air. Of the 23,402 points identified as non-air, the corresponding pixels
in the scanner images were misclassified as air 2,109 times for the high-resolution
scanner, and 2,300 times for the office scanner. When used within the calculations, a
higher void frequency will result in a lower spacing factor than is actually present. An
erroneously low spacing factor resulting from the automated analysis could be
problematic.
The use of accuracy assessment procedures for the comparison of digital images to
reference data has been in common practice in the field of remote sensing for over twenty
years [7]. However, the practice has not been widely adopted in the field of automated
air void analysis. Instead, researchers have relied upon inter-laboratory testing to
compare manual and automated methods [6]. Inter-laboratory testing is very important
for the establishment of variability in results between labs and methods, but does little to
establish the accuracy of the methods. Since there is no reference material for ASTM C
457, it is impossible to determine the true accuracy of any method [1, 8]. With the
absence of a reference material, the best that can be done is a direct point to pixel
comparison between the manual and automated method. The advantage of a direct point
to pixel comparison is that systematic differences between the automated and manual
methods can be identified, and attributed as potential sources of the variability between
the methods.
It is believed that accuracy of the void frequency discerned by the scanner methods
described here can be improved by adding additional image processing steps to the script
program. Other researchers have employed digital filters, such as erosion and dilation, or
cut-off filters, to eliminate isolated pixels identified as air [3, 6]. Additionally, the ability
of the scanners themselves to achieve increased resolution will only continue to improve
with technological advances. Further research will explore these avenues while
continuing to refine the application of the flat-bed scanner on ASTM C 457.
References
1)
C457-98 Standard Test Method for Microscopical Determination of Parameters of
the Air-Void System in Hardened Concrete, American Society for Testing and
Materials, West Conshohocken, Pennsylvania, 2000.
2)
Dewey G.R., Darwin D., “Image Analysis of Air Voids In Air-Entrained
Concrete” Structural Engineering and Engineering Materials SM Report No. 29,
University of Kansas Center for Research, August 1991.
3)
Pade C., Jakobsen U.H., Elsen J., “A New Automatic Analysis System for
Analyzing the Air Void System in Hardened Concrete.” Proceedings of the 24th
International Conference on Cement Microscopy, San Diego, California, pp. 204213, 2002.
4)
Peterson K.W., Swartz R.A., Sutter L.L., Van Dam T.J., “Air Void Analysis of
Hardened Concrete with a Flatbed Scanner” Proceedings of the 24th International
Conference on Cement Microscopy, San Diego, California, pp. 304-316, 2002.
5)
Sutter L.L., “Macro Programming with NIH Image for Implementing ASTM
C457” Proceedings of the 20th International Conference on Cement Microscopy,
Guadalajara, Mexico, pp. 382-384, 1998.
6)
Zhang Z., Ansari F., Vitillo N., “Automated Determination of Entrained Air-Void
Parameters in Hardened Concrete.” ACI Materials Journal, V. 102, No. 1,
January-February 2005.
7)
Congalton, R.G., Green, K., Assessing the Accuracy of Remotely Sensed Data:
Principles and Practices, Lewis Publishers, New York, 137 p., 1999.
8)
Saucier, F., Pleau, R., and Vezina, D., "Precision of the Air Void Characteristics
Measurement by ASTM C 457: Results of an Interlaboratory Test Program."
Canadian Journal Civ. Eng., V. 23, pp. 1118-1128, 1996.