Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

A Hybrid Image Enhancement Algorithm for Effective Concrete

Surface Crack Classification

Sheerin Sitara Noor Mohamed1 & Kavitha Srinivasan2

1
Research Scholar, Department of Computer Science and Engineering, Sri
Sivasubramaniya Nadar College of Engineering Kalavakkam, Tamil Nadu, India
2
Assosiate Professor, Department of Computer Science and Engineering, Sri
Sivasubramaniya Nadar College of Engineering Kalavakkam, Tamil Nadu, India
1
[email protected]; [email protected]

Abstract: Huge number of images are acquired and analysed every day for a range of
applications in civil infrastructure. One such application is the identification of cracks in
concrete surface images, which is a challenge owing to their low contrast and resolution,
blurriness, noise and information loss. Existing image enhancement algorithms improve
either contrast or resolution to a rather limited extent. This paper proposes a Hybrid Image
Enhancement (HIE) algorithm to improve both the contrast and resolution of concrete
surface images using the Wavelet transform and Singular Value Decomposition (SVM). The
enhanced concrete surface crack images are classified into specific crack types. The
classification comprises preprocessing, crack detection, feature extraction and crack
classification. The images are initially preprocessed using the Wiener filter to remove
blurriness, following which cracks are detected using morphological operations and
discontinuities in the segmented crack regions eliminated using the K-Dimensional Tree
algorithm. Features are extracted from the segmented regions using statistical and
geometric features. The image is classified thereafter into specific crack types using
algorithms from three different neural network, kernel and tree based categories. The
proposed HIE algorithm is validated using quantitative metrics and the results obtained
are compared with those from State-of-the-Art methods and datasets. The results have
shown that the HIE algorithm offers significantly improved accuracy of between 6% and
10% in the classification of concrete surface images.

Keywords: Concrete surface images, Singular Value Decomposition, Hybrid Image

Enhancement, K-Dimensional Tree, Types of cracks

1. Introduction

A crack is a complete or incomplete separation of a region into two or more parts,

created through aging, breaking or fracturing. Cracks are broadly classified into two
categories, regular and irregular. Regular cracks include the longitudinal [1], transverse
[2], thin [3], sealed [4], tiny [5] and large [5], while miscellaneous [1], complex [6] and
mixed [7] constitute irregular cracks. Cracks arise across different concrete, sheet metal,


Corresponding author

Volume 23, Issue 9, September - 2021 Page-1282

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

alloy steel and asphalt surfaces. Of these, cracks on concrete surfaces commonly occur in
buildings, bridges, roads and pavements, given the wear and tear associated with daily life.
In addition, concrete surfaces are used in civil infrastructure construction like foundations,
staircases and slabs because of their durability, fire resistance and cost effectiveness.
A crack in a concrete surface is indicative of degradation in the structure over a
period of time, culminating in severe environmental damage and calling for periodic
maintenance. Real time issues arising from the presence of cracks have included, in recent
times, increased road accidents1, building collapses2 and the closure of bridges3 spanning
across rivers. The occurrence of such events in real time suggests a need for the regular
maintenance of concrete surfaces and rectification of cracks at the earliest. In general,
cracks are manually checked by civil infrastructure inspectors. There are, however, issues
with such checks, including the relatively small number of human experts in the field,
inadequate domain knowledge, the time and expense entailed in inspection as well as the
risk to life involved in the careful examination of damaged structures. To address these
issues, an automatic crack detection and classification method evolved over a period, even
though many real time issues or challenges are not resolved.
This paper addresses a key issue related to the automatic crack detection and
classification of low contrast and high resolution images of cracked concrete. A Hybrid
Image Enhancement (HIE) algorithm is proposed to overcome existing drawbacks and
improve the accuracy and reliability of the image classification system for concrete cracks.
In addition, the proposed algorithm is suitable for concrete images with multiple cracks,
acquired through the use of Unmanned Aerial Vehicles (UAV), robots and low quality
sensor cameras.
The rest of this paper is organized as follows: Section 2 discusses the background to
preprocessing, image processing and machine learning in the crack detection domain. The
proposed system is explained with algorithms in the next section. Section 3 explains the
proposed system with algorithms. Section 4 describes the dataset and analyzes the
experimental results using quantitative metrics for the three datasets used. Section 5
concludes the paper with future work.

2. Related Work

Research on crack detection and classification methods began to emerge from 2002
onwards. Methods that analyze images of concrete cracks in different surfaces incorporate
preprocessing, enhancement, segmentation and classification of crack types. Popular and
widely used preprocessing techniques for cracked concrete images are the Savitzky-Golay
[8] and Wiener filters [2]. The Savitzky-Golay filter increases the signal-to-noise ratio
without distorting the image, while the Wiener filter method has fewer computations and
better noise resistance. A review of image enhancement algorithms used in the literature on
cracked concrete images in different surfaces show a preference for the Canny edge [2],
Laplace operator and histogram equalization [9], threshold approach [10], Wavelet
transform [2], [11], and Singular Value Decomposition (SVD) [12]. [2] applied the Canny
edge detection algorithm to extract structural information from images and dramatically
reduce the quantum of data to be processed. [9] suggested that the Laplace operator and

1
https://www.gilmanbedigian.com/accidents-caused-by-poor-road-quality-and-
conditions-in-dc/
2
https://www.thehindu.com/news/cities/Delhi/ashok-vihar-building-collapse-
accused-ignored-cracks/article25063214.ece
3
https://metro.co.uk/2020/08/13/major-london-bridge-closed-immediate-effect-
heatwave-causes-cracks-13125914/

Volume 23, Issue 9, September - 2021 Page-1283

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

histogram equalization methods reduce noise and enhance image contrast. [10] applied the
thresholding approach for enhanced contrast in concrete surface images prior to detecting
regions with cracks. [11] applied the statistical characteristics of the wavelet coefficient in
different frequency bands to maximize the resolution of road surface images. [12] applied
the SVD on low contrast asphalt surface images to enhance them prior to detecting regions
with prominent cracks. [13] proposed a hybrid method (a combination of the min-max gray
level discrimination, Otsu method and shape analysis algorithm) to identify crack defects
with better accuracy by boosting the contrast of building surface images. It is inferred from
these studies that there is no single algorithm which improves contrast and resolution
simultaneously for concrete surface images. Therefore, a HIE algorithm using a
combination of the Discrete Wavelet Transform (DWT) and SVD is proposed to refine the
contrast and resolution of concrete surface images before classifying crack types. [14]
suggested the same combination in their review of concrete surface image classification
approaches. Further, it is observed that the combined DWT and SVD approach works just
as well on medical images by enhancing the brightness of CT scan images [15].
Crack regions are identified, following which the region of interest is segmented using
crack detection techniques such as least square binarization [16], Otsu method, integrated
algorithm [17] and gradient method [18]. The crack region is identified and the region of
interest segmented using maximum entropy [16], Histogram of Oriented Gradients (HOG)
[19], morphological operations [2, 17, 20 - 21]. Maximum entropy and least square
binarization are combined to sustain the uniform grey values within each block using block
wise binarization. The HOG method reduces the influence of illumination information and
identifies cracked regions by capturing silhouettes and contour information. The Otsu
method performs automatic image thresholding to distinguish the foreground from the
background, based on the single intensity value. A morphological operation processes the
image based on shapes, using dilation, erosion and reconstruction approaches. This
particular method is chosen in this research, since the boundary regions are correctly
identified and the pattern of the crack is retrieved.
From the segmented crack region, significant features are extracted for crack classification.
The different features extracted include the mean, standard deviation, area, eccentricity,
length, width, correlation, rectangularity length, aspect ratio length, intensity and histogram
[1-2, 17, 21]. However, [22] proved that while a large number of significant features offers
higher accuracy, the computational time is likely to increase, based on the application.
The extracted features are used to classify cracks into specific longitudinal,
transverse, crocodile and non-crack types, using supervised algorithms like the Support
Vector Machine (SVM), Multi-Layer Perceptron (MLP), AdaBoost, random forest, and
decision tree [2, 10, 21, 24 - 25]. Of these, the SVM is ideally suited to small sample sets
because of its strong generalization ability which requires that the number of dimensions
be greater than the number of samples for effectiveness [2]. [21] suggested that the MLP
obtains rapid prediction for large datasets owing to its feed forward design, even though it
takes longer time for training. [5] found that the AdaBoost is flexible enough to be
combined with other machine learning algorithms to enhance performance, though the
dataset has noisy data and outliers. A study by [24] found that the random forest is robust
to noise and most suitable for large datasets with a minimum possibility of overfitting. [23]
suggested that the decision tree is suited to both classification and regression problems
because it accepts incomplete data and supports numerical and categorical features. [25]
suggested that the decision tree heuristic algorithm, which combines the C4.5 and J48, is
preferable for multi-crack classification. The dataset used in this paper is considered for
validating the proposed HIE algorithm using machine learning techniques.
Development in technology and computations using a Graphics Processing Unit
(GPU) have facilitated the introduction of Deep Neural Network (DNN) frameworks for
image classification. The DNN techniques applied in crack image classification include
Convolutional Neural Networks (CNN) and transfer learning approaches using pre-trained
models. [26] compared the CNN with the SVM and boosting method for binary crack
classification, resulting in 89%, 73% and 74% accuracy, respectively. Also, the dataset used

Volume 23, Issue 9, September - 2021 Page-1284

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

in [26] is considered for validating the proposed HIE algorithm using machine learning
techniques. Further, [27] suggested that the enhanced crack images improved their
performance by 1% to 12% with the CNN and pre-trained VGG16, VGG19 and Inception
ResNet models. Images without enhancement, on the contrary, showed little improvement,
thus justifying the importance of enhancement algorithms for crack images.
From the study of existing work on crack detection and classification, appropriate methods
are chosen and modified for the preprocessing, enhancement, segmentation and
classification of the crack region. Also, a hybrid approach is proposed to overcome issues
with low resolution and contrast through enhancement.

3. System Design

The design of the proposed system, using the HIE algorithm for the effective crack
classification of low resolution and low contrast images, is shown in Figure 1. The model
comprises two stages: (i) Crack detection and connection, and (ii) Feature extraction and
crack classification.

Figure 1. Proposed system for crack detection and classification

Volume 23, Issue 9, September - 2021 Page-1285

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

3.1. Crack Detection and Connection

3.1.1 Preprocessing - Wiener filter: The Wiener filter is used in preprocessing to

eliminate blurriness in the image. It also reduces the mean square error by maintaining the
tradeoff between noise smoothening and inverse filtering. In addition, information on tiny
cracks is preserved for accurate classification and the process described in Algorithm 1.

Volume 23, Issue 9, September - 2021 Page-1286

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

3.1.2 Hybrid Image Enhancement (HIE) – Wavelet transform and SVD: The
proposed HIE algorithm is a combination of the Wavelet transform and SVD methods.
Wavelet transformation, which is applied to consolidate information on cracks in the lower
approximation sub band, is further enhanced using the SVD. This is because SVD, enhances
the crack target by eliminating the noise inference and improving the resolution with
contrast. The proposed HIE is described in Algorithm 2 and the workflow shown in Figure
2. The DWT, EHL, ELH, EHH, ELL and IDWT in Figure 2 stand for the Discrete Wavelet
Transform, Estimated High Low, Estimated Low High, Estimated High High, Estimated
Low Low, and Inverse Discrete Wavelet Transform, respectively. The HIE algorithm works
in 3 stages: resolution enhancement (Function 1), contrast enhancement (Function 2) and
reconstruction of the enhanced image in phases (Function 3).

Volume 23, Issue 9, September - 2021 Page-1287

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

Volume 23, Issue 9, September - 2021 Page-1288

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

Figure 2. Proposed Hybrid Image Enhancement approach

3.1.3 Crack segmentation and connection - Morphological operations and KD Tree:

Crack regions are segmented from the enhanced image using appropriate morphological
operations. In this method, adjacent pixels are connected using a closing operation. Also,
the discontinuities in the crack segments are reconnected to a limited extent by filling in the
smaller segments using K-Dimensional Tree (KD Tree) explained in Algorithm 3.

3.2. Feature Extraction and Crack Classification

In this stage, features are extracted from the segmented region with cracks, which are
classified into specific types, based on their image features. In all, two statistical and six
geometric features are retrieved. The two statistical features namely L_row (number of
crack target intersecting with horizontal lines) and L_col (number of crack targets
intersecting with vertical lines). The six geometric features are area, rectangularity, aspect
ratio, eccentricity ratio, height and width. Neural network (MLP), kernel (SVM) and tree
based approaches (AdaBoost, random forest and decision tree) are used to classify crack
images from the extracted features into binary (crack vs non-crack) or multiclass regular,
irregular and non–crack types [2, 4, 28].

Volume 23, Issue 9, September - 2021 Page-1289

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

4. Experiments

The proposed algorithm for crack detection, and classification with validation, are
implemented in Matlab R2018b and Python 3.3.2, respectively, for the three datasets used.
In this section, the description of the datasets, experimental results, and a performance
evaluation with appropriate comparisons are explained.

4.1. The Dataset

For ease of explanation, existing datasets are named dataset 1 [26], dataset 2 [23]
and the dataset acquired is dataset 3, described in Table 1, identifying the type of crack as
well as the number of training and test images. Dataset 3 is a collection of images from
DIGITLab, incorporating web and crack images acquired using a mobile camera on
different concrete surfaces. Dataset 1 contains 40000 (20000 crack and 20000 non-crack)
images with mostly prominent crack regions, captured using a low cost sensor. Dataset 2

Volume 23, Issue 9, September - 2021 Page-1290

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

contains 1018 (788 crack and 230 non-crack) low resolution, low contrast and/or shadowy
images collected from different surfaces. Dataset 3 has 431 concrete (328 crack and 103
non-crack) low resolution, low contrast images. The dataset includes regular (longitudinal,
transverse, thin, line-like, medium and large) and irregular (miscellaneous, crocodile,
reflexive and complex) cracks.
Table 1. Dataset Description
Dataset Crack Type Training Set Test Set (No. Total Images
(No. of images) of images)
Dataset 1 [26] Crack 16000 4000 20000
Non-crack 16000 4000 20000
Total 32000 8000 40000
Dataset 2 [23] Crack 477 311 788
Non-crack 145 85 230
Total 788 230 1018
Binary Crack 253 75 328
class Non-crack 83 20 103
Dataset 3 Total 336 95 431
(Collected Regular 124 36 160
dataset) Multi - Crack
class Irregular 129 39 168
crack
Non-crack 83 20 103

4.2. Experimental Results

The stage wise output obtained for the input image is illustrates in Figure 3, with
WF, WE and WoE standing for the Wiener Filter, With Enhancement and Without
Enhancement. It is evident that the input image without the HIE retains noise, unlike the
crack region that is clearly segmented and classified with the HIE.
Further, the information on the enhanced crack images is analyzed with the original image
using the metrics of entropy and standard deviation. It is inferred from the results that the
entropy and standard deviation increased by 7.6% and 24.92%, respectively, for the
enhanced images depicted in Table 2. In addition, noise levels are reduced and the
appropriate feature values extracted.

Figure 3. Result of Proposed System at Every Stage

Volume 23, Issue 9, September - 2021 Page-1291

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

Table 2. Performance of the HIE using Entropy and Standard Deviation

Concrete Crack Image Entropy Standard Deviation
Original image 5.8145 16.6888
Enhanced image 6.7631 27.7702

The KD Tree algorithm is used to connect the discontinuities in the crack region,
following image enhancement and a morphological operation, as shown in Figure 4. Each
connected component in the crack region represents the crack line, and the K-Nearest
Neighbor (KNN) is used to distinguish the endpoint of each crack line, shown in different
colors. Finally, based on the minimum distance value, the crack regions are connected and
depicted in red.

Figure 4. Result of KD Tree

4.3. Performance Analysis

From the segmented crack region, eight features (described in Section 3) are
extracted and classified using the random forest, MLP, AdaBoost, SVM and decision tree
algorithms. The classification results are analysed, performance-wise, for the three datasets
(described in Table 1) using the metrics of accuracy, recall and precision. The results of the
binary (crack vs. non-crack) and multiclass (regular, irregular and non-crack) classification
are shown as graphs in Figure 5 and Figure 6, wherein each dataset with the HIE and
Without Enhancement (WoE) is depicted in green and orange, respectively, and the State-
of-the-Art (SoA) dataset in brown. The graphs shown are generated using Tableau
Software. The scale used in the graph ranges from 0.0 to 1.0, with a step size of 0.1.

Volume 23, Issue 9, September - 2021 Page-1292

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

Figure 5. Graphical representation of accuracy Vs. datasets and methods

for binary classification

The inferences obtained from an analysis of the features and the five classification
algorithms versus the datasets (1, 2 and 3) for the HIE and WoE are listed below:
(i) From the extracted features, L_row and L_col plays an efficient role in computing
characteristic difference between similar cracks at pixel level.
(ii) Dataset 1 resulted in 12% and 7% higher accuracy than dataset 2 and dataset 3,
respectively, for the HIE, given that the prominence of the crack regions in dataset 1
resulted in better classification accuracy. (iii) As far as dataset 3 is concerned, binary class
classification outperformed multiclass classification. In both cases, however, enhanced
images outperformed than without enhanced images, as shown in Figure 5 and Figure 6.
(iv) In terms of classifiers, the random forest and AdaBoost algorithms are best suited to
the classification of crack types, irrespective of the datasets (Figure 5 and Figure 6).
(v) A Comparative analysis with existing work: The proposed algorithm performed better
for dataset 1 using the random forest rather than the CNN model (SOA) [23], because more
significant and relevant features could be extracted. The metrics of accuracy, precision and
recall are used for comparison, resulted with the improved values of 7%, 9% and 4%,
respectively. For multiclass classification, the SOA [25] model resulted in a success rate of
80%, whereas the proposed model with enhanced images achieved 6% more accuracy for
dataset 3 because it incorporated the HIE approach, resulting in an improved performance
overall.

Volume 23, Issue 9, September - 2021 Page-1293

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

Figure 6. Graphical representation of accuracy Vs. datasets and methods

for multiclass classification

Table 3 depicts a performance analysis of Datasets Vs Methods for the quantitative

metrics of accuracy, precision and recall. Table 3 makes it clear that the proposed HIE
algorithm has improved the performance overall, irrespective of the dataset and
classification model used. With respect to five types of classification models considered,
the random forest has the highest accuracy for enhanced and the AdaBoost for without
enhanced crack images, even though the overall performance is less.
Table 3. Performance Analysis (Datasets Vs Methods)
Dataset Accuracy Precision Recall Method(s)
WoE 0.86 0.86 0.86 AdaBoost and Random forest
Dataset 1 HIE 0.96 0.96 0.96 Random forest
WoE 0.79 0.81 0.80 AdaBoost
Dataset 2 HIE 0.84 0.85 0.84 Random forest
WoE 0.79 0.78 0.79 Support Vector Machine
Dataset 3 HIE 0.89 0.92 0.91 AdaBoost
SOA 0.89 0.87 0.92 Convolutional Neural
Network

5. Conclusion and Future Work

This research has proposed an automatic crack detection and classification scheme
to deal with cracks on concrete surface images, and the results obtained were compared
with State-of-the-Art methods and datasets. The objective of the research was three-fold:

Volume 23, Issue 9, September - 2021 Page-1294

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

to classify low resolution or low contrast images into specific types, eliminate
discontinuities in the crack region during segmentation, and apply a suitable classification
algorithm for crack classification. The stages of the proposed system included
preprocessing, image enhancement, segmentation of the crack region and classification of
cracks into appropriate types. Preprocessing using the Wiener filter eliminated blurriness
in the crack images. Low contrast and low resolution images were improved using the
proposed HIE technique. Further, the crack regions were segmented by filling holes using
morphological operations and the discontinuities connected using the KD Tree algorithm.
Features extracted from the segmented crack region were classified into specific crack types
using the random forest, MLP, AdaBoost, SVM and decision tree algorithms. The
importance of the proposed HIE was analysed from the crack classification results for
different datasets using the quantitative metrics of accuracy, precision and recall. Our
findings revealed that the classification results for low contrast, low resolution and the
blurred images shown a significant improvement with the HIE than without HIE. In
addition, the results were computed and analysed for different datasets for binary and
multiclass classification. The results of dataset 2 are more or less identical to those of the
proposed dataset 3, because the images are low resolution, low contrast and/or shadowy in
nature. Dataset 1 and 3 resulted in better classification accuracy for enhanced images in
terms of both binary and multiclass classification, thus demonstrating the efficacy of the
proposed HIE technique in civil infrastructure assessment.
The proposed work can be further improved for use in real time applications by
maintaining a distance between the mobile camera and the location of the crack, capturing
sustained high resolution images at specific time intervals through the use of UAVs or
robots, and developing a single system that works to detect and classify all surface crack
types.

Acknowledgments

We proffer our deepest gratitude to the Department of CSE, Sri Sivasubramaniya

Nadar College of Engineering, for permitting us to utilize their High Performance
Computing Laboratory to complete this research successfully. However, this research
received no specific grant from any funding agency in the public, commercial, or not-for-
profit sectors.

REFERENCES
[1] H. Oliveira and P. L. Correia, “Identifying and Retrieving Distress Images from
Road Pavement Surveys”, IEEE International Conference on Image Processing,
San Diego, USA, (2008) October 12-15.
https://doi.org/10.1109/icip.2008.4711690
[2] Y. Chen, T. Mei, X. Wang and F. Li, F, “A Bridge Crack Image Detection and
Classification Method based on Climbing Robot”, Thirty fifth Chinese Control
Conference, Chengdu, China, (2016) July 28-30.
https://doi.org/10.1109/chicc.2016.7553984
[3] Y. S. Yang, C. M. Yang and C. W. Huang, C. W, “Thin Crack Observation in a
Reinforced Concrete Bridge Pier Test using Image Processing and Analysis”,
Advances in Engineering Software, vol. 83, (2015), pp. 99-108.
https://doi.org/10.1016/j.advengsoft.2015.02.005
[4] M. Kamaliardakani, L. Sun and M. Ardakani, “Sealed-Crack Detection
Algorithm using Heuristic Thresholding Approach”, International Journal of
Computing in Civil Engineering, vol. 30, no. 1, (2016), pp. 1-10.
https://doi.org/10.1061/(asce)cp.1943-5487.0000447
[5] P. Shi, X. Fan, J. Ni, Z. Khan and M. Li, “A Novel Underwater Dam Crack
Detection and Classification Approach based on Sonar Images”, Sarah and

Volume 23, Issue 9, September - 2021 Page-1295

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

George Journals, vol. 12, no. 6, (2017), pp. 541-554.

https://doi.org/10.1371/journal.pone.0179627
[6] W. Li, H. Ju, S. Tighe, Q. Ren and Z. Sun, “Three-Dimensional Pavement Crack
Detection Algorithm based on Two-Dimensional Empirical Mode
Decomposition”, Journal of Transportation Engineering, vol. 143, no. 2, (2017),
pp. 1-12. https://doi.org/10.1061/jpeodx.0000006
[7] M. Salman, S. Mathavan, K. Kamal and M. Rahman, “Pavement Crack
Detection using the Gabor Filter”, International IEEE Conference on
Intelligent Transportation Systems, Hague, Netherlands, (2013), October 6-9.
https://doi.org/10.1109/itsc.2013.6728529
[8] H. N. Nguyen, T. Y. Kam and P. Y. Cheng, “A Novel Automatic Concrete Surface
Crack Identification using Isotropic Undecimated Wavelet Transform”. IEEE
International Symposium on Intelligent Signal Processing and Communication
Systems, Tamsui, Taiwan, (2012), December 3-6.
https://doi.org/10.1109/ISPACS.2012.6473594
[9] X. Qiu, G. Li, J. Wei and C, “A Method of Crack Defects Real-time Image
Enhancement in Continuous Casting Billet”, Symposium on Photonics and
Optoelectronics, Guilin, China, (2012), October 29-31.
https://doi.org/10.1109/sopo.2012.6270558
[10] R. Amhaz, S. Chambon, J. Idier and V. Baltazart, “Automatic Crack Detection
on Two-Dimensional Pavement Images: An Algorithm based on Minimal Path
Selection”, IEEE Transactions on Intelligent Transportation Systems, vol. 17,
no. 10, (2016), pp. 2718-2729. https://doi.org/10.1109/TITS.2015.2477675
[11] X. Zhou, L. Xu and J. Wang, “Road Crack Edge Detection based on Wavelet
Transform”, Institute of Physics Publishing Conference series: Earth and
Environmental Science, vol. 237, no. 3, (2019), pp. 1-5.
https://doi.org/10.1088/1755-1315/237/3/032132
[12] M. E. Torbaghan, W. Li, N. Metje, M. Burrow, D. Chapman and C. D. F. Rogers,
“Automated Detection of Cracks in Roads using Ground Penetrating Radar”,
Journal of Applied Geophysics, vol. 179, (2020), pp. 104-118.
https://doi.org/10.1016/j.jappgeo.2020.104118
[13] N. Hoang, N, “Detection of Surface Crack in Building Structures using Image
Processing Technique with an Improved Otsu Method for Image Thresholding”,
Advances in Civil Engineering, vol. 2018, (2018), pp. 1-10.
https://doi.org/10.1155/2018/3924120
[14] N. Sheerin Sitara, S. Kavitha and G. Raghuraman, “Review and Analysis of
Crack Detection and Classification Techniques based on Crack Types”,
International Journal of Applied Engineering Research, vol. 13, no. 7, (2018),
pp. 6056-6062https://www.ripublication.com/ijaer18/ijaerv13n8_63.pdf
[15] F. Kallel and A. B. Hamida, “A New Adaptive Gamma Correction based
Algorithm using DWT-SVD for Non-Contrast CT Image Enhancement”, IEEE
Transactions on Nanobioscience, vol. 16, no. 8, (2017), pp. 666-675.
https://doi.org/10.1109/tnb.2017.2771350
[16] M. Qiao, W. Xiaoying, W and Z. Yuan, “Research on a Least Squares
Thresholding Algorithm for Pavement Crack Detection”, Sixth International
Conference on Information Science and Technology, London, UK, (2016), July
26-28. https://doi.org/10.1109/icist.2016.7483459
[17] P. Wang and H. Huang, “Comparison Analysis on Present Image-based Crack
Detection Methods in Concrete Structures”, Third International Congress on
Image and Signal Processing, Vancouver, Canada, (2010), August 26-28.
https://doi.org/10.1109/cisp.2010.5647496
[18] H. Liu, Z. Huang, Y. Zang and A. M. H. Ahmed, “Shape-based Micro Crack
Detection in Plastic through Image Processing”, International Journal of
Signal Processing, Image Processing and Pattern Recognition, vol. 9, no. 1,
(2016), pp. 281-288. https://doi.org/10.14257/ijsip.2016.9.1.27

Volume 23, Issue 9, September - 2021 Page-1296

Journal of University of Shanghai for Science and Technology ISSN: 1007-6735

[19] L. Meng, Z. Wang and F. Y. Fujikawa, “Detecting Cracks on a Concrete Surface

using Histogram of Oriented Gradients”, International Conference on
Advanced Mechatronic Systems, Beijing, China, (2015), August 22-24.
https://doi.org/10.1109/icamechs.2015.7287137
[20] P. Prasanna, K. J. Dana, N. Gucunski, B.B. Basily, H. M. La, R. S. Lim and H.
Parvardeh, “Automated Crack Detection on Concrete Bridges”, IEEE
Transactions on Automation Science and Engineering, vol. 13, no. 2, (2016),
pp. 591-599. https://doi.org/10.1109/tase.2014.2354314
[21] W. Zhang, Z. Zhang, D. Qi and Y. Liu,. “Automatic Crack Detection and
Classification Method for Subway Tunnel Safety Monitoring”, Sensors, vol. 14,
no. 10, (2014), pp. 19307-19328. https://doi.org/10.3390/s141019307
[22] P. Chun, S. Izumi and T. Yamane, “Automatic Detection Method of Cracks from
Concrete Surface Imagery using Two-Step Light Gradient Boosting Machine”,
Computer Aided Civil and Infrastructure Engineering, vol. 36, no. 1, (2020),
pp. 61-72. https://doi.org/10.1111/mice.12564
[23] M. Maguire, S. Dorafshan and R. J. Thomas,” SDNET2018: An annotated
Image Dataset for Non-Contact Concrete Crack Detection using Deep
Convolutional Neural Networks”, Data in Brief, vol. 21, (2018), pp. 1664-
1668. https://doi.org/10.1016/j.dib.2018.11.015
[24] H. Oliveira and P. L. Correia, “Automatic Road Crack Detection and
Characterization”, IEEE Transactions on Intelligent Transportation Systems,
vol. 14, no. 1, (2013), pp. 155-168. https://doi.org/10.1109/TITS.2012.2208630
[25] A. Cubero Fernandez, F. J. Rodriguez-Lozano, R. Villatoro, J. Olivares and J.
M. Palomares, “Efficient Pavement Crack Detection and Classification”,
EURASIP Journal on Image and Video Processing, vol. 39, (2017), pp. 1-11.
https://doi.org/10.1186/s13640-017-0187-0
[26] N. M. Zaki, S. Deris, S and K. K. Chin, “Extending the Decomposition Algorithm
for Support Vector Machines Training”. Journal of Information and
Communication Technology, vol. 1, no. 2, (2002), pp. 17-29. http://e-
journal.uum.edu.my/index.php/jict/article/view/7815
[27] N. Sheerin Sitara and S. Kavitha, “Comparative Analysis of Deep Neural
Networks for Crack Image Classification”, International Conference on
Intelligent Data Communication Technologies and Internet of Things, Tamil
Nadu, India (2019), August 7-8. https://doi.org/10.1007/978-3-030-34080-
3_49
[28] A. Mohan and S. Poobal, “Crack Detection using Image Processing: A Critical
Review and Analysis”, Alexandria Engineering Journal, vol. 57, no. 2, (2018),
pp. 787-798. https://doi.org/10.1016/j.aej.2017.01.020

Volume 23, Issue 9, September - 2021 Page-1297