SlideShare a Scribd company logo
1 of 10
Download to read offline
Wavelet-based Asphalt Concrete Texture Grading and
Classification
Ali Almuntashri, Sos Agaian
Department of Electrical and Computer Engineering, University of Texas at San Antonio
San Antonio, TX, USA 78249
ABSTRACT
In this Paper, we introduce a new method for evaluation, quality control, and automatic grading of texture images
representing different textural classes of Asphalt Concrete (AC). Also, we present a new asphalt concrete texture
grading, wavelet transform, fractal, and Support Vector Machine (SVM) based automatic classification and recognition
system. Experimental results were simulated using different cross-validation techniques and achieved an average
classification accuracy of 91.4.0 % in a set of 150 images belonging to five different texture grades.
Keywords: texture classification, asphalt concrete grading, fractal dimension, wavelet features, statistical classification
1. INTRODUCTION
Texture analysis is a significant image processing tool that provides essential information and characteristics about
images. Texture perception plays an important role in Human Visual System (HVS) perception for interpretation and
recognition purposes 1
. Texture analysis and classification is considered to be a challenging task and of a key importance
2
.The processing of texture images provides information used in wide areas of applications ranging from medical
applications to natural scenes. Examples of image processing applications include image segmentation, image
categorization, image registration, image visualization, and pattern recognition.
Texture can be defined as the set of local neighborhood properties of gray levels of a region in an image 2
. Literature
have proposed different approaches for textures classification using variety of methods, such as co-occurrence matrix,
second and higher order statistics, frequency transform and multi resolution techniques. Recently, multiresolution
techniques such as Discrete Wavelet Transform (DWT) and Gabor Transform are applied in a wider scale for texture
classification 3.
In this paper, focus is given to the classification of texture images that belong to Stone Matrix Asphalt (SMA) in Asphalt
Concrete (AC) mixtures. Such texture images offer a rich field of texture analysis and characterization, and have a
significant role in civil engineering applications. Asphalt concrete is composed of uniquely complex heterogeneous
materials of air voids, mastics, and aggregates. Mastics consist of binder and fines. The overall performance of AC is
highly dependent of the proportions of these materials as well as the distribution and characterization of their physical
properties 4,5
.The development of high resolution X-ray computed tomography (CT) has shown a considerable promise
to efficiently characterize the AC microstructure. X-ray CT imaging technique generates 2D and 3D high resolution
texture images with the capability of capturing the details of microstructures. Several studies have demonstrated the
potential application of such imaging technique to characterize different properties of AC mixtures. Recently, it is used
effectively to quantify air void distributions, aggregate orientation, segregation, and surface textures 5,6
.
In 2005, the green highways initiative started with a significant objective which is to foster partnerships for improving
upon the natural, built, and social environmental conditions in watershed areas, while sustaining life-cycle functional
requirements of transportation infrastructure.16
In the USA, most highways were built in 1950s and 1960s and currently
have a significant deterioration. More than 96% of the current highways are covered asphalt pavement. 16
Image Processing: Algorithms and Systems IX, edited by Jaakko T. Astola, Karen O. Egiazarian,
Proc. of SPIE-IS&T Electronic Imaging, SPIE Vol. 7870, 78701A · © 2011 SPIE-IS&T
CCC code: 0277-786X/11/$18 · doi: 10.1117/12.876682
SPIE-IS&T/ Vol. 7870 78701A-1
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
These numbers give a significant indication about the future needs for asphalt pavement materials such as aggregates. In
the same time, conformal to the green highways initiative is a mandatory objective 16
. One of the main solutions is the
use of recycled pavement materials in asphalt concrete mixture. Hence, a proper way for quality control, verification,
classification, and grading of asphalt concrete mixture becomes of an important task.
In this work, we propose a grading and a classification system to analyze and grade asphalt concrete components by
using texture classification algorithms. This paper is organized as follows: a description of proposed grading standard for
asphalt concrete mixtures based on their textural properties is illustrated in section 2. Also, a brief background on
wavelet transform and our previous work in RGB analysis is given in section 3. In Section 4, the proposed algorithm is
discussed. Section 5 discusses simulation results along with different simulation examples. Finally, a conclusion is
presented in Section 6.
2. GRADING OF ASPHALT CONCRETE TEXTURE PATTERNS
Manual grading of asphalt concrete mixtures is based on human visual perception. This approach is very subjective and
subject to intra and inter-observer variations. Also, it is a time consuming task and raises difficulties as far as spatial
resolution is considered especially in the subgroups of homogenous and semi-homogenous classes. To overcome these
issues, automatic classification of asphalt concrete classes is needed especially with the existence of advanced computing
power and effective image processing algorithms.
In this paper, we combine features from wavelet transform to provide a new features extraction approach in the
automatic classification of 150 images that belong to five textural grades of asphalt concrete images (figure 1). Up to our
knowledge, there is no existing standard grading system for asphalt concrete mixture. Hence, we propose these major
five grades based to represent the highly dominant textures in asphalt concrete mixtures based on our observations. Also,
there may exist other sub-grades that may fall in between any two major grades.
These grades are chosen based on their textural complexity of asphalt concrete components as follows:
Grade 1: this grade is illustrated in the previous figure of the AC image on the left side of the figure. It has the most
homogenous texture compared to other grades. Mastics are the dominant components of the AC mixture. The second
predominant components are represented by the aggregates which are hard to distinguish visually. Air voids represent
the least AC components in this grade.
Grade 2: this grade (second from the left) has a less homogenous, or a semi-homogenous AC mixture compared to the
previous grade. Air gaps patterns are spread more than the previous grade. Also, the presence of aggregates is more
dominant than the previous grade.
Figure 1 The proposed five grades of asphalt concrete texture images (Grade 1-Grade 5 from left to right)
SPIE-IS&T/ Vol. 7870 78701A-2
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
Grade 3: T
Aggregates r
larger textura
Grade 4: gra
become harde
Grade 5: the
start to take p
in this grade o
3.1 Wav
Wavelet Tran
( )
,
x
m n
ψ us
For the const
two-scale diff
( ) 2
x
k
φ = ∑
For wavelet d
be given as:
1,
j n
c + = ∑
For 2-D wave
and high pass
signal up to tw
Horizontal, V
This pattern (m
epresent the d
al space as the
ade4 texture (
er to be visual
main feature
place compare
of AC images
velet Transfo
nsform: In w
ing translation
truction of the
fference equati
( ) (2 )
h k x k
φ −
decomposition
, ( 2
j k
k
c h k −
∑
elet transform
s filters, h and
wo levels of d
Vertical, and D
middle image
dominant com
mastics.
( fourth from
lly observed. A
of this grade (
ed to previous
s.
orm
wavelet transf
n and dilation
, ( )
m n x
ψ =
e mother func
ion expressed
) (2) ; ψ
n to a Jth-leve
1,
) ; j n
n d +
m, the basis fu
d g, respective
decomposition
Diagonal subm
F
e) is more of
mponents of th
m left) has al
Also, more he
(last from left
s patterns. Sm
3.
form, the sig
n of the mother
/ 2
2 (2
m m
ψ
− −
=
ction ( )
x
ψ , the
as:
( ) 2
k
x
ψ = ∑
el, the relation
, (
j k
k
c g k
= ∑
unctions opera
ly. The follow
ns. The coeffic
matrices in the
Figure 2 1-D w
f a heterogen
he AC mixtur
lmost equal p
eterogeneity is
t) is the domin
mall aggregate
BACKGR
nal is transfo
r wavelet (x
ψ
)
x n
− (1)
e determinatio
( ) (2
k
g k x
φ
∑
n of the decom
2 )
n
− (5-
ate along the h
wing figure ill
cients at the H
e 2D transform
wavelet decompo
eous mixture
re. Also, air v
portions of ag
s observed due
nance of the a
s and air void
ROUND
ormed into a
)
x as follows:
for integers
on of a scalin
)
k
− (3)
mposed coeffi
-6) where 0
horizontal and
lustrates the w
HH, HG, GH, a
m space.
osition filter ban
compared to
voids are easi
ggregates, mas
e to the smalle
ir voids. Larg
ds reserve alm
a family of o
s m and n
ng function (
φ
where ( )
g k
icients to direc
0 j J
≤ ≤
d vertical dire
wavelet decom
and GG corre
nk
o the previou
ily observed r
stics and air
er aggregates
ger gaps betwe
most the same
orthogonal ba
( )
x is needed
) ( 1) (1
k
h k
= − −
ct previous co
ctions as a pa
mposition filter
spond to the A
us two grades
representing a
gaps. Mastics
size.
een aggregates
textural space
ases functions
d to satisfy the
) (4)
k
oefficients can
air of low pass
r bank for 1-D
Approximated
s.
a
s
s
e
s
e
n
s
D
d,
SPIE-IS&T/ Vol. 7870 78701A-3
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
3.2 Previous work in RGB morphological filtering
Color textural images need to be pre- processed in order to investigate the contribution of color components in textural
patterns. This step is mandatory to be incorporated into texture analysis and classification. Color distribution over a
textural space can be described by the relationship between chromatic and structural distribution 13
. Achromatic color
component is used for textural pattern information. However, luminance (Y), and chrominance ( U and V) components
are utilized for extracting color information 13,14,15
. Considering HIS color model, the pure color is described by the hue
(H) component, while the degree of the color dilution by white light is described by the saturation (S) component13
. In
RGB model, RGB color model consists of three independent color planes representing the three chromaticities
components: Red, Green, and Blue. In order to specify a pixel RGB value, each color component is specified
independently within the range 0 to 255. 7
In 7
, we presented an algorithm that focuses on extracting edge details that cannot be highlighted for detection using the
standard assigned HVS vector weight for RGB colors. Given a tri-chromatic RGB image, one might consider designing a
rational morphological filter that takes into consideration the assigned HVS weight for each component in tri-chromatic
space. Let Fn and Fm denote two different as follows:
1 2 1 3
2 1 2 3
2 1 2 3
: (0.2989R,0.5870G,0.1140B) ; (7)
: ( log (0.2989 ), log (0.5870 ), log (0.1140 ));
( log ( ), log ( ));
( log ( ), log ( ));
( log ( ), log ( )) (8)
Fn
R G B
Fm R R G G B B
R R
R w R w G R w R w B
G G
G w G w R G w G w B
B B
B w B w R B w B w G
α α α
β β β
α α
β β
α α
β β
α α
β β
− −
− −
− −
where β and α are constants assigned to each color channel, 1 2 3
, ,
w w w represent different assigned weight for R,G,B
color components respectively. The ratio of different morphological operations may take any of the following forms:
max( )
1
max( )
Fn
Fm
β
⎛ ⎞
⎜ ⎟
⎝ ⎠
;
min( )
2
max( )
Fn
Fm
β
⎛ ⎞
⎜ ⎟
⎝ ⎠
;
min( )
3
min( )
Fn
Fm
β
⎛ ⎞
⎜ ⎟
⎝ ⎠
;
max( )
4
min( )
Fn
Fm
β
⎛ ⎞
⎜ ⎟
⎝ ⎠
(9) ; where β is a normalization constant
The above expressions yield successful results in investigating the intensity of each RGB component at each pixel taking
into consideration the HVS perception characteristics. This algorithm is used as a preprocessing step for colored textural
images prior to features extraction and classification.
4. ALGORITHM
4.1 Haar wavelet energy feature
In our work, we consider Haar wavelet decomposition for feature vectors generation in wavelet domain. The Haar
wavelet transform is an orthogonal transform that provides a transform domain where differential energy is concentrated
in localized areas. Haar transform is considered to be one of the simplest and the fastest wavelet transform especially
when considering the fast-Haar transform 8
. The Haar transform basis function can be expressed as follows:
SPIE-IS&T/ Vol. 7870 78701A-4
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
(11)
Considering features extraction from wavelet domain, the major extracted features in wavelet analysis are wavelet
energy and entropy due to their popularity in features extraction and classification applications 9
. In this paper, we use
the wavelet energy feature “F1”, that can be expressed as follows:
2
ij
i j
x
energy
nxn
=
∑∑
(12)
4.2 Phase measure of wavelet coefficients
Wavelet coefficients of a one dimensional signal can be expressed as :
1 1 2 1 2
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) (13)
j
i j
i
x n A n D n A n D n D n
D n A n
= + = + +
= +
∑
where “D” represents the detailed coefficients, and “A” represents approximated coefficients.
In order to investigate phase difference information between approximated and detailed coefficients, we investigate the
following relation in a 2D representation:
1
tan (14)
ij
i j
ij
i j
A
D
θ −
⎛ ⎞
⎜ ⎟
= ⎜ ⎟
⎜ ⎟
⎠
⎝
∑∑
∑∑
This relation represents the second extracted feature “F2” in Haar wavelet domain. However, prior to utilizing phase
feature vector in the final classification procedure, it is normalized using the least-square regression normalization.
1 1
(2) (10)
1 1
H
+ +
⎛ ⎞
= ⎜ ⎟
+ −
⎝ ⎠
1
(2 ) ( 1 1)
[ ] (2 ) , 2,3, ,
1 1
2 2 (2 ) ( 1 1)
n
H
n
Haar H n
n n n
I
−
⎛ ⎞
⊗ + +
⎜ ⎟
= = =
⎜ ⎟
− − ⊗ + −
⎝ ⎠
K
SPIE-IS&T/ Vol. 7870 78701A-5
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
4.3 Wavelet-based Fractal Dimension (FD)
The architectural and geometrical features of asphalt concrete images present a rich domain for fractal analysis through
the concept of fractal dimension (FD). The fractal dimension provides a geometrical description of an image by
measuring the degree of complexity and how much space a given set occupies around each of its points 10
. The
variations of different architectures and components of AC images associated with different grades shows distinct
features that can be recognized for classification using fractal dimension. In this part, the concept of deriving fractal
dimension from wavelet decomposed detailed coefficients is presented.
One might consider the detailed coefficients in a given wavelet decomposition as a domain for representing visual
textural characteristics in terms of horizontal, vertical and diagonal details. Wavelet coefficients are first mapped to the
spatial domain of graylevel scale in the bounded range of [0,255]. Then, a further threshold stage for binarization
purpose is introduced as an adaptive process to reflect local statistics by deploying standard deviation in local windows
11
. The proposed binarization threshold T is expressed as follows:
( , ) ( , ) . ( , ) (15)
T i j mean i j C i j
σ
= +
Where ( , )
mean i j is the local mean, C is a global constant, and ( , )
i j
σ is the local standard deviation.
The purpose of the previous processes is to have an image with edge details suitable for fractal analysis using fractal
dimension (FD) measure. Given a set S in Euclidean n-space, the self similarity of S is obtained if Nr distinct copies of S
exist by r ratio 12
. The following equation provides the fractal dimension measure “FD” as the slope measure expressed
as follows:
log( )
(16)
log(1/ )
N
r
FD
r
=
For the submatrices of the horizontal, vertical, and diagonal architectural details, the fractal dimension is found by
averaging all slopes/fractal dimensions from each detailed submatrix as follows:
, ,
( )
(17)
H V D
i
D n
i
FD
i n
= ∑
where FD (n) represents the average fractal dimension of all detailed submatrices (n) ,at decomposition level i. For
features normalization, the generated feature vector is normalized using the least-squared regression method. This
procedure of using wavelet-based fractal measure for generating the third feature vector “F3” will be combined with the
previous two feature vectors for the final classification presented in the next section. The following block diagram
illustrates the overall procedure of the developed algorithm.
SPIE-IS&T/ Vol. 7870 78701A-6
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
5. SIMULATION RESULTS AND DISCUSSION
The data set used in simulation consists of 150 textural images of asphalt concrete mixture. These images are taken from
asphalt core using X-ray Computed Tomography (CT) scan. The data set is divided into 5 subgroups, with 30 images
each according to their textural complexities, as shown previously in figure 1. Images were prepared by the University of
Texas at El Paso 5
.
5.1 Wavelet energy
In order to investigate the performance of Haar wavelet energy feature, the Support Virtual Machine (SVM) classifier
with the general “Hold-Out” cross validation method is used. Haar wavelet decomposition is performed at two
decomposition levels on the normalized features of the five grades with 30 images in each grade. The classification
performance is presented in the following table.
Table 1 Classification performance of Haar wavelet energy feature
Grades
Energy feature vector “F1”
Correct Classification %
Decomposition
Level 1
Decomposition
Level2
Grade 1 86.6 86.6
Grade 2 80.0 80.0
Grade 3 80.0 96.0
Grade 4 94.6 93.3
Grade 5 80 80.0
Average 84.2 87.1
Based on the observation of classification performance from the previous table, it is noted that a better classification
performance is achieved using the energy feature at the second decomposition level. Also, there is a superior
classification performance of grade 3 textures compared to the remaining grades. This increases the average
Texture
Image
F1=Energy
F2=Phase
F3=FD
A D1
“H”
D2
“V”
D3
“D”
Feature
vectors fitting
and
Normalization
SVM
Classifier
Figure 3 Block diagram of the proposed algorithm
Haar wavelet
Decomposition
RGB pre-
processing
SPIE-IS&T/ Vol. 7870 78701A-7
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
classification performance using the second decomposition level. Hence, decomposition level 2 is chosen for generating
the “Haar” energy feature vector, denoted as “F1”, prior to the final SVM classification for all feature vectors.
5.2 Wavelet Phase Measure
In this part, a feature vector is generated using the phase of wavelet approximated coefficients ratio to the sum of all
detailed coefficients as expressed previously. The generated feature vector “F2” is normalized before the SVM classifier
is utilized to measure the classification performance. The following table illustrates simulation results using this feature.
Table 2 Classification performance using wavelet phase measure
Grades
Phase feature
vector “F2”
Correct
Classification %
Grade 1 80.0
Grade 2 80.0
Grade 3 80.0
Grade 4 81.3
Grade 5 89.3
Average 82.3
The phase information-based classification resulted in a higher performance in grade 5 textures compared to the previous
wavelet energy feature. Although a lower average is achieved, the final combination of features is expected to result in a
higher classification performance due to the effect of the increase classification performance for grade 5 texture images.
5.3 Wavelet-based Fractal Dimension (FD)
In this part, the generated feature vector “F3” represents the average fractal dimension measure for all the detailed
coefficients. The purpose of utilizing such feature vector is to highlight the similarities of textural architectures observed
visually at the submatrices of the detailed coefficients. The results of the normalized feature vector classification using
SVM are indicated in the below table.
Table 3 Classification performance using wavelet-based fractal measure
Grades
FD feature
vector “F3”
Correct
Classification %
Grade 1 80.0
Grade 2 78.8
Grade 3 80.0
Grade 4 97.3
Grade 5 81.2
Average 83.46
SPIE-IS&T/ Vol. 7870 78701A-8
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
As can be observed from this table above, the major advantage of utilizing wavelet-based fractal dimension is to have a
higher discrimination of grade 4 texture images. The next step is to fuse all features for classification which is
highlighted in the next section using different cross validation methods.
5.4 Features Combination
The generated feature vectors F1, F2 and F3 are combined for classification using SVM classifier with the following
different cross validation procedures: Hold-out all samples, Leave-one-out, K-fold cross validation with k=5 and 10. In
K-fold cross validation, feature selection is performed K times, where each time it is performed on a different training
set. From the table below, the highest correct classification rate of 91.4% is achieved using 5-fold cross validation
method. The lowest classification rate achieved is 80% using the “leave-one-out” technique.
Table 4 Overall classification performance
Texture
Class
F1+F2+F3
Hold-out
F1+F2+F3
Leave-One-Out
F1+F2+F3
5-fold
F1+F2+F3
10-fold
Grade 1 94.6
Average
88.8%
100.0
Average
80.0%
97.3
Average
91.4%
98.0
Average
91.2%
Grade 2 84.0 50.0 79.3 78.6
Grade 3 82.6 100.0 88.0 86.6
Grade 4 96.0 100.0 96.6 96.6
Grade 5 86.6 50. 96.0 96.0
6. CONCLUSION
In this Paper, we introduce a new method for evaluation, quality control, and verification of asphalt concrete mixtures
using automatic grading of their texture images. Also, we presented a new novel asphalt concrete texture grading system
based on textural patterns. The introduced algorithms combine features from wavelet transform and wavelet-based
fractal analysis. The small number of extracted features proved to be effective and superior to other classification
algorithms utilizing larger sets of feature vectors where a dimensionality problem occurs. Moreover, the proposed
classification system has a fast performance due to the selection of Haar wavelet transform for features extraction, and
SVM classifier for feature vectors classification. The proposed algorithm reached 91.4% of correct classification rate,
which indicates its effectiveness in textural classification problems.
SPIE-IS&T/ Vol. 7870 78701A-9
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
REFERENCES
[1]. Laine, A., Fan, J., “Textures classification by wavelet packet signatures”, IEEE Transactions on pattern analysis and
machine intelligence, 15(11), 1186-1191(1993).
[2]. Livens, S., Scheunders, P., Van De Wouwer, G., Van Dyck, D., “Wavelets for texture analysis”, Proc. IEEE of the
sixth international conference on image processing and its application”, 2, 581-585(1997).
[3]. Arivazhagan, S., Ganesan, L., “Texture segmentation using wavelet transform”, Pattern recognition letters, 24,
3197–3203(2003).
[4]. Zelelew, H. M., Papagiannakis, A. T., Masad, E., "Application of digital image processing techniques for asphalt
concrete mixture images" The 12th international conference of international association for computer methods and
advanced geomachines (IACMAG). Goa, India, (2008).
[5]. Zelelew, H. M., Papagiannakis, A. T., "A volumetric thresholding algorithm for processing asphalt concrete x-ray
CT images". International journal of pavement engineering, (2007).
[6]. Agaian, S., Almuntashri, A., Papagiannakis, A.T., “An improved Canny edge detection application to asphalt
concrete images”, Proc. IEEE of the international conference on system, man, and cybernetics, 3683-3687(2009).
[7]. Almuntashri, A., Agaian, S., “An algorithm for visualizing and detecting edges in RGB color images using
logarithmic ratio approach” Proc. IEEE SMC(2010), 3942-3947(2010)
[8]. Agaian, S., Tourshan, K., Noona, J. P., “Parameterization of Salnt-Haar transform”, Proc. IEEE on vision, image and
signal processing, 150(5), 306-312(2003).
[9]. Sengur, A., “Color texture classification using wavelet transform and neural network ensembles”, The Arabian
journal for science and engineering, 34(2B), 491-502(2009).
[10]. Tang, Y. Y., Tao, Y., Lam, E. C. M., “New method for feature extraction based on fractal behavior”, Pattern
Recognition, 35, 1071–1081 (2002).
[11]. Barney Smith, E., Likforman-Sulem, L., and Darbon, J., "Effect of pre-processing on binarization", Proc. SPIE,
(7534), document recognition and retrieval XVII (2010).
[12]. Sarkar, N., Chaudhuri, B. B., “An efficient differential box-counting approach to compute fractal dimension of
image”, IEEE Transactions on Systems, Man and Cybernetics24(1), 115-120(1994).
[13]. Arivazhagan, S., Ganesan, L., Angayarkanni, V. “ Color texture classification using wavelet transform”, Proc.
IEEE of the sixth international conference on computational intelligence and multimedia applications, 315-320(2005).
[14]. Paschos, G., Valavanis, K., “A color texture based visual monitoring system for automated surveillance”, IEEE
Transaction on systems, man and cybernetics, (29)1, 298-306(1999).
[15]. Jain, A., Healy, G., “A multiscale representation including opponent color features for texture recognition,” IEEE
Transactions on Image Processing, 7, 124-128(1998).
[16]. Edil, T. B., “Green Highways: strategy for recycling materials for sustainable construction practices”, Seventh
International Congress on Advances in Civil Engineering, Turkey, 1-20(2006)
SPIE-IS&T/ Vol. 7870 78701A-10
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

More Related Content

Similar to 12.876682.pdf

Kentaro_region_filling_inpainting
Kentaro_region_filling_inpaintingKentaro_region_filling_inpainting
Kentaro_region_filling_inpainting
Vipin Gupta
 
Barber_TU2.T03_hk_mb_casa_fg_hk_FINAL.pptx
Barber_TU2.T03_hk_mb_casa_fg_hk_FINAL.pptxBarber_TU2.T03_hk_mb_casa_fg_hk_FINAL.pptx
Barber_TU2.T03_hk_mb_casa_fg_hk_FINAL.pptx
grssieee
 
A full experimental and numerical modelling of the practicability of thin foa...
A full experimental and numerical modelling of the practicability of thin foa...A full experimental and numerical modelling of the practicability of thin foa...
A full experimental and numerical modelling of the practicability of thin foa...
Mehran Naghizadeh
 
Lect 6 pavement design
Lect 6 pavement designLect 6 pavement design
Lect 6 pavement design
M Firdaus
 

Similar to 12.876682.pdf (20)

A Numerical Method for Modelling Discontinuous Mechanics of Asphalt Mixture
A Numerical Method for Modelling Discontinuous Mechanics of Asphalt MixtureA Numerical Method for Modelling Discontinuous Mechanics of Asphalt Mixture
A Numerical Method for Modelling Discontinuous Mechanics of Asphalt Mixture
 
xfem_3DAbaqus.pdf
xfem_3DAbaqus.pdfxfem_3DAbaqus.pdf
xfem_3DAbaqus.pdf
 
Kentaro_region_filling_inpainting
Kentaro_region_filling_inpaintingKentaro_region_filling_inpainting
Kentaro_region_filling_inpainting
 
Journal EASTS 2015
Journal EASTS 2015Journal EASTS 2015
Journal EASTS 2015
 
An Experimental Study on Vehicular Emission Dispersion through Single Storied...
An Experimental Study on Vehicular Emission Dispersion through Single Storied...An Experimental Study on Vehicular Emission Dispersion through Single Storied...
An Experimental Study on Vehicular Emission Dispersion through Single Storied...
 
COMPUTATION OF HARMONIC GREEN’S-FUNCTIONS OF A HOMOGENEOUS SOIL USING AN AXIS...
COMPUTATION OF HARMONIC GREEN’S-FUNCTIONS OF A HOMOGENEOUS SOIL USING AN AXIS...COMPUTATION OF HARMONIC GREEN’S-FUNCTIONS OF A HOMOGENEOUS SOIL USING AN AXIS...
COMPUTATION OF HARMONIC GREEN’S-FUNCTIONS OF A HOMOGENEOUS SOIL USING AN AXIS...
 
30120140504021 2
30120140504021 230120140504021 2
30120140504021 2
 
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
 
20320140505007
2032014050500720320140505007
20320140505007
 
IRJET- Structural Analysis of Transmission Tower: State of Art
IRJET-  	  Structural Analysis of Transmission Tower: State of ArtIRJET-  	  Structural Analysis of Transmission Tower: State of Art
IRJET- Structural Analysis of Transmission Tower: State of Art
 
Jgrass-NewAge: Kriging component
Jgrass-NewAge: Kriging componentJgrass-NewAge: Kriging component
Jgrass-NewAge: Kriging component
 
Barber_TU2.T03_hk_mb_casa_fg_hk_FINAL.pptx
Barber_TU2.T03_hk_mb_casa_fg_hk_FINAL.pptxBarber_TU2.T03_hk_mb_casa_fg_hk_FINAL.pptx
Barber_TU2.T03_hk_mb_casa_fg_hk_FINAL.pptx
 
A full experimental and numerical modelling of the practicability of thin foa...
A full experimental and numerical modelling of the practicability of thin foa...A full experimental and numerical modelling of the practicability of thin foa...
A full experimental and numerical modelling of the practicability of thin foa...
 
Mechanical Response Analysis of Asphalt Pavement Structure
Mechanical Response Analysis of Asphalt Pavement StructureMechanical Response Analysis of Asphalt Pavement Structure
Mechanical Response Analysis of Asphalt Pavement Structure
 
Lect 6 pavement design
Lect 6 pavement designLect 6 pavement design
Lect 6 pavement design
 
D1082833
D1082833D1082833
D1082833
 
International Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentInternational Journal of Engineering Research and Development
International Journal of Engineering Research and Development
 
Applications of layered theory for the analysis of flexible pavements
Applications of layered theory for the analysis of flexible pavementsApplications of layered theory for the analysis of flexible pavements
Applications of layered theory for the analysis of flexible pavements
 
1 viscoelastic foundation on soil under vertical load
1 viscoelastic foundation on soil under vertical load1 viscoelastic foundation on soil under vertical load
1 viscoelastic foundation on soil under vertical load
 
CONSISTENT AND LUMPED MASS MATRICES IN DYNAMICS AND THEIR IMPACT ON FINITE EL...
CONSISTENT AND LUMPED MASS MATRICES IN DYNAMICS AND THEIR IMPACT ON FINITE EL...CONSISTENT AND LUMPED MASS MATRICES IN DYNAMICS AND THEIR IMPACT ON FINITE EL...
CONSISTENT AND LUMPED MASS MATRICES IN DYNAMICS AND THEIR IMPACT ON FINITE EL...
 

Recently uploaded

Recently uploaded (20)

UGC NET Paper 1 Mathematical Reasoning & Aptitude.pdf
UGC NET Paper 1 Mathematical Reasoning & Aptitude.pdfUGC NET Paper 1 Mathematical Reasoning & Aptitude.pdf
UGC NET Paper 1 Mathematical Reasoning & Aptitude.pdf
 
ICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptxICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptx
 
This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.
 
Magic bus Group work1and 2 (Team 3).pptx
Magic bus Group work1and 2 (Team 3).pptxMagic bus Group work1and 2 (Team 3).pptx
Magic bus Group work1and 2 (Team 3).pptx
 
Introduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsIntroduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The Basics
 
Mehran University Newsletter Vol-X, Issue-I, 2024
Mehran University Newsletter Vol-X, Issue-I, 2024Mehran University Newsletter Vol-X, Issue-I, 2024
Mehran University Newsletter Vol-X, Issue-I, 2024
 
General Principles of Intellectual Property: Concepts of Intellectual Proper...
General Principles of Intellectual Property: Concepts of Intellectual  Proper...General Principles of Intellectual Property: Concepts of Intellectual  Proper...
General Principles of Intellectual Property: Concepts of Intellectual Proper...
 
Spatium Project Simulation student brief
Spatium Project Simulation student briefSpatium Project Simulation student brief
Spatium Project Simulation student brief
 
Key note speaker Neum_Admir Softic_ENG.pdf
Key note speaker Neum_Admir Softic_ENG.pdfKey note speaker Neum_Admir Softic_ENG.pdf
Key note speaker Neum_Admir Softic_ENG.pdf
 
Food safety_Challenges food safety laboratories_.pdf
Food safety_Challenges food safety laboratories_.pdfFood safety_Challenges food safety laboratories_.pdf
Food safety_Challenges food safety laboratories_.pdf
 
Holdier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdfHoldier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdf
 
Asian American Pacific Islander Month DDSD 2024.pptx
Asian American Pacific Islander Month DDSD 2024.pptxAsian American Pacific Islander Month DDSD 2024.pptx
Asian American Pacific Islander Month DDSD 2024.pptx
 
Application orientated numerical on hev.ppt
Application orientated numerical on hev.pptApplication orientated numerical on hev.ppt
Application orientated numerical on hev.ppt
 
ComPTIA Overview | Comptia Security+ Book SY0-701
ComPTIA Overview | Comptia Security+ Book SY0-701ComPTIA Overview | Comptia Security+ Book SY0-701
ComPTIA Overview | Comptia Security+ Book SY0-701
 
microwave assisted reaction. General introduction
microwave assisted reaction. General introductionmicrowave assisted reaction. General introduction
microwave assisted reaction. General introduction
 
Dyslexia AI Workshop for Slideshare.pptx
Dyslexia AI Workshop for Slideshare.pptxDyslexia AI Workshop for Slideshare.pptx
Dyslexia AI Workshop for Slideshare.pptx
 
psychiatric nursing HISTORY COLLECTION .docx
psychiatric  nursing HISTORY  COLLECTION  .docxpsychiatric  nursing HISTORY  COLLECTION  .docx
psychiatric nursing HISTORY COLLECTION .docx
 
Grant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingGrant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy Consulting
 
Python Notes for mca i year students osmania university.docx
Python Notes for mca i year students osmania university.docxPython Notes for mca i year students osmania university.docx
Python Notes for mca i year students osmania university.docx
 
PROCESS RECORDING FORMAT.docx
PROCESS      RECORDING        FORMAT.docxPROCESS      RECORDING        FORMAT.docx
PROCESS RECORDING FORMAT.docx
 

12.876682.pdf

  • 1. Wavelet-based Asphalt Concrete Texture Grading and Classification Ali Almuntashri, Sos Agaian Department of Electrical and Computer Engineering, University of Texas at San Antonio San Antonio, TX, USA 78249 ABSTRACT In this Paper, we introduce a new method for evaluation, quality control, and automatic grading of texture images representing different textural classes of Asphalt Concrete (AC). Also, we present a new asphalt concrete texture grading, wavelet transform, fractal, and Support Vector Machine (SVM) based automatic classification and recognition system. Experimental results were simulated using different cross-validation techniques and achieved an average classification accuracy of 91.4.0 % in a set of 150 images belonging to five different texture grades. Keywords: texture classification, asphalt concrete grading, fractal dimension, wavelet features, statistical classification 1. INTRODUCTION Texture analysis is a significant image processing tool that provides essential information and characteristics about images. Texture perception plays an important role in Human Visual System (HVS) perception for interpretation and recognition purposes 1 . Texture analysis and classification is considered to be a challenging task and of a key importance 2 .The processing of texture images provides information used in wide areas of applications ranging from medical applications to natural scenes. Examples of image processing applications include image segmentation, image categorization, image registration, image visualization, and pattern recognition. Texture can be defined as the set of local neighborhood properties of gray levels of a region in an image 2 . Literature have proposed different approaches for textures classification using variety of methods, such as co-occurrence matrix, second and higher order statistics, frequency transform and multi resolution techniques. Recently, multiresolution techniques such as Discrete Wavelet Transform (DWT) and Gabor Transform are applied in a wider scale for texture classification 3. In this paper, focus is given to the classification of texture images that belong to Stone Matrix Asphalt (SMA) in Asphalt Concrete (AC) mixtures. Such texture images offer a rich field of texture analysis and characterization, and have a significant role in civil engineering applications. Asphalt concrete is composed of uniquely complex heterogeneous materials of air voids, mastics, and aggregates. Mastics consist of binder and fines. The overall performance of AC is highly dependent of the proportions of these materials as well as the distribution and characterization of their physical properties 4,5 .The development of high resolution X-ray computed tomography (CT) has shown a considerable promise to efficiently characterize the AC microstructure. X-ray CT imaging technique generates 2D and 3D high resolution texture images with the capability of capturing the details of microstructures. Several studies have demonstrated the potential application of such imaging technique to characterize different properties of AC mixtures. Recently, it is used effectively to quantify air void distributions, aggregate orientation, segregation, and surface textures 5,6 . In 2005, the green highways initiative started with a significant objective which is to foster partnerships for improving upon the natural, built, and social environmental conditions in watershed areas, while sustaining life-cycle functional requirements of transportation infrastructure.16 In the USA, most highways were built in 1950s and 1960s and currently have a significant deterioration. More than 96% of the current highways are covered asphalt pavement. 16 Image Processing: Algorithms and Systems IX, edited by Jaakko T. Astola, Karen O. Egiazarian, Proc. of SPIE-IS&T Electronic Imaging, SPIE Vol. 7870, 78701A · © 2011 SPIE-IS&T CCC code: 0277-786X/11/$18 · doi: 10.1117/12.876682 SPIE-IS&T/ Vol. 7870 78701A-1 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 2. These numbers give a significant indication about the future needs for asphalt pavement materials such as aggregates. In the same time, conformal to the green highways initiative is a mandatory objective 16 . One of the main solutions is the use of recycled pavement materials in asphalt concrete mixture. Hence, a proper way for quality control, verification, classification, and grading of asphalt concrete mixture becomes of an important task. In this work, we propose a grading and a classification system to analyze and grade asphalt concrete components by using texture classification algorithms. This paper is organized as follows: a description of proposed grading standard for asphalt concrete mixtures based on their textural properties is illustrated in section 2. Also, a brief background on wavelet transform and our previous work in RGB analysis is given in section 3. In Section 4, the proposed algorithm is discussed. Section 5 discusses simulation results along with different simulation examples. Finally, a conclusion is presented in Section 6. 2. GRADING OF ASPHALT CONCRETE TEXTURE PATTERNS Manual grading of asphalt concrete mixtures is based on human visual perception. This approach is very subjective and subject to intra and inter-observer variations. Also, it is a time consuming task and raises difficulties as far as spatial resolution is considered especially in the subgroups of homogenous and semi-homogenous classes. To overcome these issues, automatic classification of asphalt concrete classes is needed especially with the existence of advanced computing power and effective image processing algorithms. In this paper, we combine features from wavelet transform to provide a new features extraction approach in the automatic classification of 150 images that belong to five textural grades of asphalt concrete images (figure 1). Up to our knowledge, there is no existing standard grading system for asphalt concrete mixture. Hence, we propose these major five grades based to represent the highly dominant textures in asphalt concrete mixtures based on our observations. Also, there may exist other sub-grades that may fall in between any two major grades. These grades are chosen based on their textural complexity of asphalt concrete components as follows: Grade 1: this grade is illustrated in the previous figure of the AC image on the left side of the figure. It has the most homogenous texture compared to other grades. Mastics are the dominant components of the AC mixture. The second predominant components are represented by the aggregates which are hard to distinguish visually. Air voids represent the least AC components in this grade. Grade 2: this grade (second from the left) has a less homogenous, or a semi-homogenous AC mixture compared to the previous grade. Air gaps patterns are spread more than the previous grade. Also, the presence of aggregates is more dominant than the previous grade. Figure 1 The proposed five grades of asphalt concrete texture images (Grade 1-Grade 5 from left to right) SPIE-IS&T/ Vol. 7870 78701A-2 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 3. Grade 3: T Aggregates r larger textura Grade 4: gra become harde Grade 5: the start to take p in this grade o 3.1 Wav Wavelet Tran ( ) , x m n ψ us For the const two-scale diff ( ) 2 x k φ = ∑ For wavelet d be given as: 1, j n c + = ∑ For 2-D wave and high pass signal up to tw Horizontal, V This pattern (m epresent the d al space as the ade4 texture ( er to be visual main feature place compare of AC images velet Transfo nsform: In w ing translation truction of the fference equati ( ) (2 ) h k x k φ − decomposition , ( 2 j k k c h k − ∑ elet transform s filters, h and wo levels of d Vertical, and D middle image dominant com mastics. ( fourth from lly observed. A of this grade ( ed to previous s. orm wavelet transf n and dilation , ( ) m n x ψ = e mother func ion expressed ) (2) ; ψ n to a Jth-leve 1, ) ; j n n d + m, the basis fu d g, respective decomposition Diagonal subm F e) is more of mponents of th m left) has al Also, more he (last from left s patterns. Sm 3. form, the sig n of the mother / 2 2 (2 m m ψ − − = ction ( ) x ψ , the as: ( ) 2 k x ψ = ∑ el, the relation , ( j k k c g k = ∑ unctions opera ly. The follow ns. The coeffic matrices in the Figure 2 1-D w f a heterogen he AC mixtur lmost equal p eterogeneity is t) is the domin mall aggregate BACKGR nal is transfo r wavelet (x ψ ) x n − (1) e determinatio ( ) (2 k g k x φ ∑ n of the decom 2 ) n − (5- ate along the h wing figure ill cients at the H e 2D transform wavelet decompo eous mixture re. Also, air v portions of ag s observed due nance of the a s and air void ROUND ormed into a ) x as follows: for integers on of a scalin ) k − (3) mposed coeffi -6) where 0 horizontal and lustrates the w HH, HG, GH, a m space. osition filter ban compared to voids are easi ggregates, mas e to the smalle ir voids. Larg ds reserve alm a family of o s m and n ng function ( φ where ( ) g k icients to direc 0 j J ≤ ≤ d vertical dire wavelet decom and GG corre nk o the previou ily observed r stics and air er aggregates ger gaps betwe most the same orthogonal ba ( ) x is needed ) ( 1) (1 k h k = − − ct previous co ctions as a pa mposition filter spond to the A us two grades representing a gaps. Mastics size. een aggregates textural space ases functions d to satisfy the ) (4) k oefficients can air of low pass r bank for 1-D Approximated s. a s s e s e n s D d, SPIE-IS&T/ Vol. 7870 78701A-3 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 4. 3.2 Previous work in RGB morphological filtering Color textural images need to be pre- processed in order to investigate the contribution of color components in textural patterns. This step is mandatory to be incorporated into texture analysis and classification. Color distribution over a textural space can be described by the relationship between chromatic and structural distribution 13 . Achromatic color component is used for textural pattern information. However, luminance (Y), and chrominance ( U and V) components are utilized for extracting color information 13,14,15 . Considering HIS color model, the pure color is described by the hue (H) component, while the degree of the color dilution by white light is described by the saturation (S) component13 . In RGB model, RGB color model consists of three independent color planes representing the three chromaticities components: Red, Green, and Blue. In order to specify a pixel RGB value, each color component is specified independently within the range 0 to 255. 7 In 7 , we presented an algorithm that focuses on extracting edge details that cannot be highlighted for detection using the standard assigned HVS vector weight for RGB colors. Given a tri-chromatic RGB image, one might consider designing a rational morphological filter that takes into consideration the assigned HVS weight for each component in tri-chromatic space. Let Fn and Fm denote two different as follows: 1 2 1 3 2 1 2 3 2 1 2 3 : (0.2989R,0.5870G,0.1140B) ; (7) : ( log (0.2989 ), log (0.5870 ), log (0.1140 )); ( log ( ), log ( )); ( log ( ), log ( )); ( log ( ), log ( )) (8) Fn R G B Fm R R G G B B R R R w R w G R w R w B G G G w G w R G w G w B B B B w B w R B w B w G α α α β β β α α β β α α β β α α β β − − − − − − where β and α are constants assigned to each color channel, 1 2 3 , , w w w represent different assigned weight for R,G,B color components respectively. The ratio of different morphological operations may take any of the following forms: max( ) 1 max( ) Fn Fm β ⎛ ⎞ ⎜ ⎟ ⎝ ⎠ ; min( ) 2 max( ) Fn Fm β ⎛ ⎞ ⎜ ⎟ ⎝ ⎠ ; min( ) 3 min( ) Fn Fm β ⎛ ⎞ ⎜ ⎟ ⎝ ⎠ ; max( ) 4 min( ) Fn Fm β ⎛ ⎞ ⎜ ⎟ ⎝ ⎠ (9) ; where β is a normalization constant The above expressions yield successful results in investigating the intensity of each RGB component at each pixel taking into consideration the HVS perception characteristics. This algorithm is used as a preprocessing step for colored textural images prior to features extraction and classification. 4. ALGORITHM 4.1 Haar wavelet energy feature In our work, we consider Haar wavelet decomposition for feature vectors generation in wavelet domain. The Haar wavelet transform is an orthogonal transform that provides a transform domain where differential energy is concentrated in localized areas. Haar transform is considered to be one of the simplest and the fastest wavelet transform especially when considering the fast-Haar transform 8 . The Haar transform basis function can be expressed as follows: SPIE-IS&T/ Vol. 7870 78701A-4 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 5. (11) Considering features extraction from wavelet domain, the major extracted features in wavelet analysis are wavelet energy and entropy due to their popularity in features extraction and classification applications 9 . In this paper, we use the wavelet energy feature “F1”, that can be expressed as follows: 2 ij i j x energy nxn = ∑∑ (12) 4.2 Phase measure of wavelet coefficients Wavelet coefficients of a one dimensional signal can be expressed as : 1 1 2 1 2 ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) (13) j i j i x n A n D n A n D n D n D n A n = + = + + = + ∑ where “D” represents the detailed coefficients, and “A” represents approximated coefficients. In order to investigate phase difference information between approximated and detailed coefficients, we investigate the following relation in a 2D representation: 1 tan (14) ij i j ij i j A D θ − ⎛ ⎞ ⎜ ⎟ = ⎜ ⎟ ⎜ ⎟ ⎠ ⎝ ∑∑ ∑∑ This relation represents the second extracted feature “F2” in Haar wavelet domain. However, prior to utilizing phase feature vector in the final classification procedure, it is normalized using the least-square regression normalization. 1 1 (2) (10) 1 1 H + + ⎛ ⎞ = ⎜ ⎟ + − ⎝ ⎠ 1 (2 ) ( 1 1) [ ] (2 ) , 2,3, , 1 1 2 2 (2 ) ( 1 1) n H n Haar H n n n n I − ⎛ ⎞ ⊗ + + ⎜ ⎟ = = = ⎜ ⎟ − − ⊗ + − ⎝ ⎠ K SPIE-IS&T/ Vol. 7870 78701A-5 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 6. 4.3 Wavelet-based Fractal Dimension (FD) The architectural and geometrical features of asphalt concrete images present a rich domain for fractal analysis through the concept of fractal dimension (FD). The fractal dimension provides a geometrical description of an image by measuring the degree of complexity and how much space a given set occupies around each of its points 10 . The variations of different architectures and components of AC images associated with different grades shows distinct features that can be recognized for classification using fractal dimension. In this part, the concept of deriving fractal dimension from wavelet decomposed detailed coefficients is presented. One might consider the detailed coefficients in a given wavelet decomposition as a domain for representing visual textural characteristics in terms of horizontal, vertical and diagonal details. Wavelet coefficients are first mapped to the spatial domain of graylevel scale in the bounded range of [0,255]. Then, a further threshold stage for binarization purpose is introduced as an adaptive process to reflect local statistics by deploying standard deviation in local windows 11 . The proposed binarization threshold T is expressed as follows: ( , ) ( , ) . ( , ) (15) T i j mean i j C i j σ = + Where ( , ) mean i j is the local mean, C is a global constant, and ( , ) i j σ is the local standard deviation. The purpose of the previous processes is to have an image with edge details suitable for fractal analysis using fractal dimension (FD) measure. Given a set S in Euclidean n-space, the self similarity of S is obtained if Nr distinct copies of S exist by r ratio 12 . The following equation provides the fractal dimension measure “FD” as the slope measure expressed as follows: log( ) (16) log(1/ ) N r FD r = For the submatrices of the horizontal, vertical, and diagonal architectural details, the fractal dimension is found by averaging all slopes/fractal dimensions from each detailed submatrix as follows: , , ( ) (17) H V D i D n i FD i n = ∑ where FD (n) represents the average fractal dimension of all detailed submatrices (n) ,at decomposition level i. For features normalization, the generated feature vector is normalized using the least-squared regression method. This procedure of using wavelet-based fractal measure for generating the third feature vector “F3” will be combined with the previous two feature vectors for the final classification presented in the next section. The following block diagram illustrates the overall procedure of the developed algorithm. SPIE-IS&T/ Vol. 7870 78701A-6 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 7. 5. SIMULATION RESULTS AND DISCUSSION The data set used in simulation consists of 150 textural images of asphalt concrete mixture. These images are taken from asphalt core using X-ray Computed Tomography (CT) scan. The data set is divided into 5 subgroups, with 30 images each according to their textural complexities, as shown previously in figure 1. Images were prepared by the University of Texas at El Paso 5 . 5.1 Wavelet energy In order to investigate the performance of Haar wavelet energy feature, the Support Virtual Machine (SVM) classifier with the general “Hold-Out” cross validation method is used. Haar wavelet decomposition is performed at two decomposition levels on the normalized features of the five grades with 30 images in each grade. The classification performance is presented in the following table. Table 1 Classification performance of Haar wavelet energy feature Grades Energy feature vector “F1” Correct Classification % Decomposition Level 1 Decomposition Level2 Grade 1 86.6 86.6 Grade 2 80.0 80.0 Grade 3 80.0 96.0 Grade 4 94.6 93.3 Grade 5 80 80.0 Average 84.2 87.1 Based on the observation of classification performance from the previous table, it is noted that a better classification performance is achieved using the energy feature at the second decomposition level. Also, there is a superior classification performance of grade 3 textures compared to the remaining grades. This increases the average Texture Image F1=Energy F2=Phase F3=FD A D1 “H” D2 “V” D3 “D” Feature vectors fitting and Normalization SVM Classifier Figure 3 Block diagram of the proposed algorithm Haar wavelet Decomposition RGB pre- processing SPIE-IS&T/ Vol. 7870 78701A-7 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 8. classification performance using the second decomposition level. Hence, decomposition level 2 is chosen for generating the “Haar” energy feature vector, denoted as “F1”, prior to the final SVM classification for all feature vectors. 5.2 Wavelet Phase Measure In this part, a feature vector is generated using the phase of wavelet approximated coefficients ratio to the sum of all detailed coefficients as expressed previously. The generated feature vector “F2” is normalized before the SVM classifier is utilized to measure the classification performance. The following table illustrates simulation results using this feature. Table 2 Classification performance using wavelet phase measure Grades Phase feature vector “F2” Correct Classification % Grade 1 80.0 Grade 2 80.0 Grade 3 80.0 Grade 4 81.3 Grade 5 89.3 Average 82.3 The phase information-based classification resulted in a higher performance in grade 5 textures compared to the previous wavelet energy feature. Although a lower average is achieved, the final combination of features is expected to result in a higher classification performance due to the effect of the increase classification performance for grade 5 texture images. 5.3 Wavelet-based Fractal Dimension (FD) In this part, the generated feature vector “F3” represents the average fractal dimension measure for all the detailed coefficients. The purpose of utilizing such feature vector is to highlight the similarities of textural architectures observed visually at the submatrices of the detailed coefficients. The results of the normalized feature vector classification using SVM are indicated in the below table. Table 3 Classification performance using wavelet-based fractal measure Grades FD feature vector “F3” Correct Classification % Grade 1 80.0 Grade 2 78.8 Grade 3 80.0 Grade 4 97.3 Grade 5 81.2 Average 83.46 SPIE-IS&T/ Vol. 7870 78701A-8 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 9. As can be observed from this table above, the major advantage of utilizing wavelet-based fractal dimension is to have a higher discrimination of grade 4 texture images. The next step is to fuse all features for classification which is highlighted in the next section using different cross validation methods. 5.4 Features Combination The generated feature vectors F1, F2 and F3 are combined for classification using SVM classifier with the following different cross validation procedures: Hold-out all samples, Leave-one-out, K-fold cross validation with k=5 and 10. In K-fold cross validation, feature selection is performed K times, where each time it is performed on a different training set. From the table below, the highest correct classification rate of 91.4% is achieved using 5-fold cross validation method. The lowest classification rate achieved is 80% using the “leave-one-out” technique. Table 4 Overall classification performance Texture Class F1+F2+F3 Hold-out F1+F2+F3 Leave-One-Out F1+F2+F3 5-fold F1+F2+F3 10-fold Grade 1 94.6 Average 88.8% 100.0 Average 80.0% 97.3 Average 91.4% 98.0 Average 91.2% Grade 2 84.0 50.0 79.3 78.6 Grade 3 82.6 100.0 88.0 86.6 Grade 4 96.0 100.0 96.6 96.6 Grade 5 86.6 50. 96.0 96.0 6. CONCLUSION In this Paper, we introduce a new method for evaluation, quality control, and verification of asphalt concrete mixtures using automatic grading of their texture images. Also, we presented a new novel asphalt concrete texture grading system based on textural patterns. The introduced algorithms combine features from wavelet transform and wavelet-based fractal analysis. The small number of extracted features proved to be effective and superior to other classification algorithms utilizing larger sets of feature vectors where a dimensionality problem occurs. Moreover, the proposed classification system has a fast performance due to the selection of Haar wavelet transform for features extraction, and SVM classifier for feature vectors classification. The proposed algorithm reached 91.4% of correct classification rate, which indicates its effectiveness in textural classification problems. SPIE-IS&T/ Vol. 7870 78701A-9 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
  • 10. REFERENCES [1]. Laine, A., Fan, J., “Textures classification by wavelet packet signatures”, IEEE Transactions on pattern analysis and machine intelligence, 15(11), 1186-1191(1993). [2]. Livens, S., Scheunders, P., Van De Wouwer, G., Van Dyck, D., “Wavelets for texture analysis”, Proc. IEEE of the sixth international conference on image processing and its application”, 2, 581-585(1997). [3]. Arivazhagan, S., Ganesan, L., “Texture segmentation using wavelet transform”, Pattern recognition letters, 24, 3197–3203(2003). [4]. Zelelew, H. M., Papagiannakis, A. T., Masad, E., "Application of digital image processing techniques for asphalt concrete mixture images" The 12th international conference of international association for computer methods and advanced geomachines (IACMAG). Goa, India, (2008). [5]. Zelelew, H. M., Papagiannakis, A. T., "A volumetric thresholding algorithm for processing asphalt concrete x-ray CT images". International journal of pavement engineering, (2007). [6]. Agaian, S., Almuntashri, A., Papagiannakis, A.T., “An improved Canny edge detection application to asphalt concrete images”, Proc. IEEE of the international conference on system, man, and cybernetics, 3683-3687(2009). [7]. Almuntashri, A., Agaian, S., “An algorithm for visualizing and detecting edges in RGB color images using logarithmic ratio approach” Proc. IEEE SMC(2010), 3942-3947(2010) [8]. Agaian, S., Tourshan, K., Noona, J. P., “Parameterization of Salnt-Haar transform”, Proc. IEEE on vision, image and signal processing, 150(5), 306-312(2003). [9]. Sengur, A., “Color texture classification using wavelet transform and neural network ensembles”, The Arabian journal for science and engineering, 34(2B), 491-502(2009). [10]. Tang, Y. Y., Tao, Y., Lam, E. C. M., “New method for feature extraction based on fractal behavior”, Pattern Recognition, 35, 1071–1081 (2002). [11]. Barney Smith, E., Likforman-Sulem, L., and Darbon, J., "Effect of pre-processing on binarization", Proc. SPIE, (7534), document recognition and retrieval XVII (2010). [12]. Sarkar, N., Chaudhuri, B. B., “An efficient differential box-counting approach to compute fractal dimension of image”, IEEE Transactions on Systems, Man and Cybernetics24(1), 115-120(1994). [13]. Arivazhagan, S., Ganesan, L., Angayarkanni, V. “ Color texture classification using wavelet transform”, Proc. IEEE of the sixth international conference on computational intelligence and multimedia applications, 315-320(2005). [14]. Paschos, G., Valavanis, K., “A color texture based visual monitoring system for automated surveillance”, IEEE Transaction on systems, man and cybernetics, (29)1, 298-306(1999). [15]. Jain, A., Healy, G., “A multiscale representation including opponent color features for texture recognition,” IEEE Transactions on Image Processing, 7, 124-128(1998). [16]. Edil, T. B., “Green Highways: strategy for recycling materials for sustainable construction practices”, Seventh International Congress on Advances in Civil Engineering, Turkey, 1-20(2006) SPIE-IS&T/ Vol. 7870 78701A-10 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/23/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx