Particle Size Analysis by Image Processing Technique: November 2016

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/313604059

Particle Size Analysis by Image Processing Technique

Conference Paper · November 2016

CITATIONS READS
0 805

5 authors, including:

Khaled Mahmood Md Raquibul Hossain


Bangladesh University of Engineering and Technology Bangladesh University of Engineering and Technology
2 PUBLICATIONS   2 CITATIONS    20 PUBLICATIONS   47 CITATIONS   

SEE PROFILE SEE PROFILE

Tanvir Manzur
Ryerson University
52 PUBLICATIONS   152 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cement Technology View project

Concrete Technology View project

All content following this page was uploaded by Khaled Mahmood on 11 February 2017.

The user has requested enhancement of the downloaded file.


1ST BANGLADESH CIVIL ENGINEERING SUMMIT

BUET-ANWAR ISPAT 1st Bangladesh Civil Engineering SUMMIT 2016


BUET, Dhaka, Bangladesh
www.cesummitbd.com

Particle Size Analysis by Image Processing Technique


Khaled M. Ehsan1, Md. Raquibul Hossain2, Tanvir Manzur3, Ansary Shohag4,
Nazifa Tabassum5

Department of Civil Engineering, Bangladesh University of Engineering & Technology, Dhaka,


1
mahmoodehsan93@gmail.com, 2raquibulhossain@ce.buet.ac.bd, 3tmanzur.buet@gmail.com,
5
nazifatabassum7@gmail.com
Department of Electrical and Electronic Engineering, Bangladesh University of Engineering &
Technology, Dhaka,
4
ansary201106154@yahoo.com.

ABSTRACT
Particle size analysis technique is important to assess the quality of granular materials in material science.
It has a wide range of application in medical image processing. Image analysis technique can provide an
advanced and non-destructive way to approximate particle size of granular materials from an image. In
this research, 2-D projected images of the particles were captured and analyzed using MATLAB 2014b, a
new autonomous technique to analyze size of particles by image processing. In this method particle size
calculation is based on centroid of the particle. This paper represents an algorithm which can be applied to
a multiple number of particles and each particle’s boundary were detected and calculated in pixel unit. A
difficulty like impulse noise is discussed and a method to overcome such difficulty is also described.

KEY WORDS: Image processing; projected area; centroid; particle size distribution.

INTRODUCTION
Classification of the size of the aggregates has huge influence in civil engineering especially in advanced
material science. Particle size is important for the depiction of quality of materials. In concrete
technology, particle size analysis is mainly done using sieve analysis. It is often difficult to perform sieve
analysis in the construction site and, as a result, construction might suffer from delay. In such cases,
image processing could be used to determine particle size distribution of aggregates. However, a reliable
and verified technique will be required which can provide true distribution of particle within a batch of
aggregates. Image processing technique have already been practiced in civil engineering (Coster et al.,
2001) specially in concrete technology (Marinoni et al., 2005; Mora et al., 2000; Yue et al., 1996).2-D
image analysis are usually based on number distribution or volume based distribution with assumptions.
Both Kwan et al. (1999) and Mora et al. (1998) have done 2-D image analysis with some assumptions to
get volumetric information of the particles. 2-D image analysis has been applied for measurement of
elongation; angularity, gradation analysis etc. by a number of researchers (Al-Rousan et al., 2007; Kwan
et al., 1999; Banta et al., 2003). Banta et al. (2003) analyzed limestone of sizes varying from 4.75-25 mm
and found satisfactory image analysis results. Considering area of the particles as a parameter Kumara et
al. (2011) developed gradation curves. Sieve analysis technique is the most common technique that is
used to access particle sizes. According to CEN (1993, 1997), axial dimensions of the particles can be
calculated by processes like hand sieving, using both rectangular and square shaped sieves. According to
Fernlund (1998), sieving does not give any dimension precisely, specially for elongated and thin particles.
However, conventional methods give us particle size distribution of a sample. Image analysis allows us to
access and analyze the size of each particles in a sample. Fernlund (1998) analyzed particle sizes ranging
from 32-64 mm of railroad aggregates. Three dimensional analysis was also done in previous research

ISBN: 978-984-34-1512-7 SE | 124


1ST BANGLADESH CIVIL ENGINEERING SUMMIT

works (Lindström., 2010; Tafesse et al., 2012) for aggregate sizes ranging from 2-20 cm. For 3-D
information, the particles were imaged for two different positions one for the largest projected area and
the other was for the smallest projected area, which made that process more time consuming. Kumara
(2012) used an image analysis technique called Image J to analyze granular particles. In this study, a new
method is proposed to analyze particle size as well as irregularities in shape of objects using “Centroid”
of the image. An algorithm has been developed to get the maximum (Dmax) and minimum (Dmin)
dimension of an aggregate in pixel quantity. A possible way has been described to analyze a sample of
aggregate. The method described in this paper can calculate the maximum and the minimum dimension of
irregular shaped objects like granular particles precisely using the position of the centroid. Probable
difficulty like impulse noise may arise and procedure to overcome such difficulties has been described in
this paper.

METHODOLOGY
Firstly, captured images were transformed into binary images from RGB images. A filter as described
below was used to remove noise from binary image. Then the boundary was traced and transformed into a
normal array. Centroid of the image was then calculated using a built-in function “regionprops”. Finally
the algorithm, as described below was applied to find out Dmax and Dmin line through centroid. Figure 1
shows the flow diagram of the technique used to analyze particles.

Read Image

Conversion RGB
to Greyscale

Graythresholding

& Noise Removal

Tracing boundary

Finding centroid
of binary image

Finding
Dmax&Dmin

Figure 1: Flow diagram showing the proposed method of analyzing particle size

SE | 125
1ST BANGLADESH CIVIL ENGINEERING SUMMIT

First image was read in RGB format shown in Figure 2(a). RGB image is a three dimensional array which
represents image pixels intensity values of three fundamental colors e.g. red, green and blue. RGB images
were then converted into grayscale images shown in Figure 2(b). Grayscale image is a two- dimensional
array which represents pixels intensity values of images within [0 255].

(a) (b)

Figure 2: (a) RGB Image (b) Grayscale Image

Grayscale image was then converted into binary image which has intensity values 0 and 1, i.e “Black”
and “White” image. Thresholding operation was done by “Graythresholding” process, a global threshold
using Otsu’s method (Mathworks., 2016).Using “graythresh” function, a global threshold value was
selected to make the segmentation process more robust. Segmentation means distinguishing background
pixels from foreground pixels. It is clearly shown in Figure 3(a) that due to improper illumination and
surface irregularities of the particle, the output image consisted of holes which interfered with the actual
centroid position of the particle boundary. In order to remove holes, a post- processing technique was
adopted to remove image pixel void, i.e. holes using “imfill” function as described by Kumara et
al.(2012) (Mathworks,2016). Particles’ images contained “Salt and Pepper Noise”, an impulse noise. To
remove noise, “medfilt2” was used. As images were sensitive to noise, “bwareaopen” function was used
for further removal of noise as suggested by Manzur et al. (2016). Figure 3(b) shows the final image after
noise removal.

Holes due to improper


illumination and rough surface

(a) (b)

Figure 3: (a) After converting binary image showing holes, (b) Removal of noise using filter and filling
holes

Then boundary tracing was done using “bwboundaries” to find outermost pixel co-ordinates. Using
“cell2mat” function, the cell array was converted into ordinary array to ease the plotting of boundary
pixel co-ordinates. In order to find out the Dmax and Dmin.,“Centroid” of the images were calculated

SE | 126
1ST BANGLADESH CIVIL ENGINEERING SUMMIT

through “regionprops” function which calculates image regional properties. It gives output image as a
vector which specifies centroid of that region (Mathworks, 2016).For binary image, it is the arithmetic
mean of all the vertical and horizontal co-ordinates within that particular boundary region. The following
equation represents the algorithm used in the function to find out centroid of a binary image.

1
xc = ∑𝑥 xc = Horizontal Co-ordinate of Centroid
|𝑁|

1
yc= |𝑁| ∑ 𝑦 yc = Vertical Co-ordinate of Centroid

N= Region of the image

Here, (x,y)ɛN.

Algorithm for Finding Dmax and Dmin


Figure 4 shows a six- step algorithm to draw lines through the centroid by dividing the boundary co-
ordinates into four portions with respect to centroid. Then the lines having maximum and minimum
dimensions were calculated.

Change co-ordinate Divide boundary co- Take ratios (y/x=R)


center to blob centroid ordinates into 4 for (Q1, R1), (Q2, R2),
Cartesian co-ordinates (Q3, R3), (Q4, R4)
(Q1, Q2, Q3, and Q4)

(1) (2) (3)

Match exact or 1 ratio Find the points of the Take each pixel length
of (R1, R3) and (R2, matched lines of the lines using
R4) Eucledian distance

(4) (5) (6)

Figure 4: Six- steps algorithm for finding Dmax & Dmin.

Output for Individual Particles


The following table (Table 1) shows the output of the images after plotting the boundary pixel co-
ordinates found from the normal array formed after extracting boundary pixel co-ordinates. The
dimensions were in pixel unit found by Eucledian distance.

ANALYSIS FOR MULTIPLE NUMBERS OF PARTICLES


The proposed method showed satisfactory results for a multiple number of objects, independent of
particles’ orientation. The algorithm developed can detect each of the particles in the image and boundary
tracing could be done for individual particles. However, proper illumination system is required for precise

SE | 127
1ST BANGLADESH CIVIL ENGINEERING SUMMIT

boundary line and reduction of noise. Figure 5 shows the RGB image and Figure 6 shows the detection
and calculations of individual particles after processing according to the method proposed. Thus, it is
evident that the proposed method can analyze each individual particles present in the image.

Table 1: Graph- plotting of boundary co-ordinates & values- Dmax and Dmin.

Output Image Dmax and Dmin Processing


time& pixel
length

Dmax=
698.9878

Dmin=324.1188

Dmax=
624.2195

Dmin=424.2004

Dmax=
445.5839

Dmin=322.0248

Dmax=
540.8974

Dmin=294.4317

SE | 128
1ST BANGLADESH CIVIL ENGINEERING SUMMIT

1 2

3 4

Figure 5: RGB image for multiple number of particles

Output Images for Sample

(1) (2)

(3) (4)

Figure 6: Detecting each particle in the image showing Dmax & Dmin.

CONCLUSION
Particle size analysis of coarse aggregates was done using 2-D projected images of the particles through
MATLAB-2014b. A new technique has been proposed to analyze particles as well as for any other
irregular shaped objects using “Centroid” of the object. A particle orientation- independent algorithm has
been developed to find and calculate the size of maximum dimension and minimum dimension through
centroid of the particles from 2-D projected images. For a multiple number of particles, the algorithm is
capable of calculating dimension of each particle individually. However, proper scaling system and
calibration system will be needed for providing actual dimensions of the particles. Illumination of the
particles is a primary controlling factor for determining sharp boundary line. Therefore, appropriate
illumination is required for precise dimension calculation for a greater number of objects.

SE | 129
1ST BANGLADESH CIVIL ENGINEERING SUMMIT

REFERENCES
Al-Rousan, T., Masad, E., Tutumluer, E., and Pan, T. (2007), Evaluation of image analysis techniques for
quantifying aggregate shape characteristics, Construction and Building Materials, Vol.21:978-990
Banta, L., Cheng, K., and Zaniewski, J. (2003), Estimation of limestone particle mass from 2D images,
Power Technology, Vol.132:184-189
CEN, 1993, Tests for geometrical properties of aggregates, European Committee for Standardization
Document CEN/TC: 54/SC6 N307E
CEN, 1997, Tests for geometrical properties of aggregates, European Committee for Standardization
Document EN 933-3
Coster, M., Chermant, J. L. (2001), Image analysis and mathematical morphology for civil engineering
materials, Cement & Concrete Composites, Vol.23:133-151
Fernlund, JMR. (1998), The effect of particle form on sieve analysis: a test by image analysis,
Engineering Geology, (50): 111-124
Kumara, GHAJJ., Hayano, K., Ogiwara, K. (2011), Fundamental study on particle size distribution of
coarse materials by image analysis, 1st international conference on Geo-technique, Construction Materials
and Environment (GEOMATE 2011), pp: 399-404
Kumara, GHAJJ., Hayano, K., Ogiwara, K. (Sept.2012), Image analysis techniques on evaluation of
particle size distribution of gravel, International Journal of GEOMATE, Vol.3, No.1(Sl.No.5): 290-297
Kwan, AKH., Mora, CF., Chan, HC. (1999), Particle shape analysis of coarse aggregate using digital
image processing, Cement and Concrete Research. Vol.29:1403-1410
Lindström, H. (2010), Rock property measurements using image processing, TRITA LWR Degree Project
10:28
Manzur, T., Ehsan, KM., Sultana, SL., and Mahmud, M. (2016), Measurement of Surface Damage
through Boundary Detection: An Approach to Assess Durability of Cementitious Composites under
Tannery Wastewater, Advances in Materials Science and Engineering, vol. 2016, Article ID 5368635, 13
pages, 2016 doi:10.1155/2016/5368635
Mathworks, April 2016, http://www.mathworks.com/help/images
Marinoni, N., Pavese, A., Foi, M., Trombino, L. (2005), Characterization of mortar morphology in thin
sections by digital image processing, Cement and Concrete Research, Vol.35:1613-1619
Mora, CF., Kwan, AKH., Chan, HC. (1998), Particle size distribution analysis of coarse aggregate using
digital image processing, Cement and Concrete Research, Vol.28 No.6: 921-932
Mora, CF., Kwan, AKH. (2000), Sphericity, shape factor, and convexity measurement of coarse aggregate
for concrete using digital image processing, Cement and Concrete Research, Vol.30: 351-358
Tafesse, S., Fernlund, JMR., Bergholm, F. (2012), Digital sieving- Matlab based 3-D image analysis,
Engineering Geology,137-138(74-84)
Yue, ZQ., Morin, I. (1996), Digital image processing for aggregate orientation in asphalt concrete
mixtures, Canadian Journal of Civil Engineering, Vol. 23: 480-489

SE | 130

View publication stats

You might also like