A Comparative Review of Image Processing Based Crack Detection Techniques on Civil Engineering Structures

Journal of Soft Computing in Civil Engineering 5-3 (2021) 58-74
How to cite this article: Zawad MRS, Zawad MFS, Rahman MA, Priyom SN. A comparative review of image processing based
crack detection techniques on civil engineering structures. J Soft Comput Civ Eng 2021;5(3):58–74.
https://doi.org/10.22115/scce.2021.287729.1325.
2588-2872/ © 2021 The Authors. Published by Pouyan Press.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Contents lists available at SCCE
Journal of Soft Computing in Civil Engineering
Journal homepage: www.jsoftcivil.com
A Comparative Review of Image Processing Based Crack
Detection Techniques on Civil Engineering Structures
M.R.S. Zawad1
, M.F.S. Zawad2*
, M.A. Rahman2
, S.N. Priyom3
1. UG Student, Department of Information and Communication Technology, Bangladesh University of Professionals
(BUP), Dhaka, Bangladesh
2. UG Student, Department of Civil Engineering, Chittagong University of Engineering and Technology (CUET),
Chattogram, Bangladesh
3. PG Student, Department of Civil Engineering, Chittagong University of Engineering and Technology (CUET),
Chattogram, Bangladesh
Corresponding author: u1501083@student.cuet.ac.bd
https://doi.org/10.22115/SCCE.2021.287729.1325
ARTICLE INFO ABSTRACT
Article history:
Received: 24 May 2021
Revised: 11 August 2021
Accepted: 14 September 2021
Crack detection and repair of the cracks in engineering structures is
essential to ensure serviceability and durability. Traditionally, cracks
are detected by the examiner's visual inspection; as a result, crack
detection and estimation of characteristics are greatly dependent on the
examiner's personal judgment, which has aided in the repair of various
structures and evaluation of the crack phenomenon in previous
decades. Due to industrial advancement, the number of engineering
structures has increased, but compared to that, expertise in the crack
detection field did not raise that level. So, a less time-consuming and
more accurate approach is needed. The image processing technique
works simultaneously to detect the cracks with their attributes. In this
context, the development of the algorithm and the implementation
procedure is also simple. But some defects such as identifying noises
as cracks and weakness in identifying micro-cracks have become
significant challenges for this technique. Unable to locate transverse
cracks in concrete structures is also a vital issue. So, to develop an
accurate method, an extensive survey on the current articles is needed.
In this paper, a critical analysis has been done on crack detection
through the image processing phenomenon and a detailed literature
review to understand the prospects of this method. From the literature
review, it was observed that a general structure of CNN-based
algorithm with camera images for crack detection could be an efficient
approach with higher accuracy.
Keywords:
Cracks;
Durability;
Noises;
Transverse cracks;
Micro-cracks.

M.R.S. Zawad et al./ Journal of Soft Computing in Civil Engineering 5-3 (2021) 58-74 59
1. Introduction
Cracks are a common phenomenon in engineering structures. Especially for concrete-based
structures, cracks can occur due to cyclic load, fatigue stress and tensile stress. The cracks on
engineering structures reduces the local stiffness and cause discontinuity of the materials [1].
Moreover, the generation of cracks and widening of it in the concrete structures decreases the
lifetime of the structure and causes corrosion of the embedded rebar inside the concrete, which
ultimately fails the structure. So, cracks in the concrete surface should be treated properly as
surface cracks are critical indicators of structural damage [2]. Fortunately, early crack detection
and prevention of cracks are possible, which prohibits financial losses and casualties.
Though, crack detection in the manual process, which is dependent on the personal justification
and judgement of the specialists, has shown acceptable performance in the past decades. But it is
mainly dependent on the experience of the examiner. With modern industrialization, as the
number of structures has increased by a significant amount. So, an alternative and more exact
detection process is needed [3].
Crack detection through image processing is a technique of surface crack detection that mainly
uses image processing-based algorithms to differentiate cracks from engineering structures
surfaces. Crack detection through image processing effectively analyzes and detects
characteristics of cracks such as crack width, length and area [4]. So, automatic crack detection
can be an alternative to manual procedures that possess more accuracy and reliability [5].
Image processing is a Non-destructive testing method that can be conducted by various technical
approaches such as (i) Ultrasonic testing, (ii) Laser-based testing, (iii) Infrared and thermal
testing and (iv) Radiographic testing [6]. Due to the simplicity and accuracy of image-based
crack detection, the interest among researchers in developing more convenient techniques is
increasing day by day. Recent studies on bridges, dams, and tall buildings also indicated that to
enhance the structure's durability, an assessment of the accurate service life and present condition
is needed. In this case, image-based assessment is more effective than traditional inspection [7].
Though image-based crack detection is a promising technique but some limitations of this
technique, such as: counting surface noises as cracks, unable to detect the direction of the
propagation of cracks properly, and limited practical use, has developed a big challenge for the
researchers to overcome and establish a more proper and accurate technique. So, a deep study of
the existing methodology is needed.
In this research study, an attempt has been taken to summarize various research findings based
on crack detection through image processing to find out the existing pros and cons and analyze
the future prospects of this technique in the field of Structural engineering.
2. Methodology
This paper was conducted by analyzing the various published research articles depending on the
method followed for image processing and the significant outcomes obtained in the research by

60 M.R.S. Zawad et al./ Journal of Soft Computing in Civil Engineering 5-3 (2021) 58-74
analyzing the key information of the published articles. 30 research articles, including scientific
journals and conference proceedings, were reviewed. Research articles were selected based on
the titles, keywords and abstracts. Fig. 1 represents the research process flow that has been
followed for conducting this review article.
Fig. 1. Methodology followed for this study.
3. Basics of image processing based crack detection
Image processing is the way of controlling image properties to analyze and extract intended
distinctive attributes from the images. Some set of rules or processes followed to extract the
attributes from an image are known as image processing algorithms. Fig. 2 resembles the general
implementation method of image processing.
Fig. 2. General implementation stages of image processing.
An image-based crack detection system has several benefits, such as large storage of data and
detection of the propagation of cracks on various engineering structure surfaces. During the
initial days of implementing the image processing technique for identifying cracks, more
emphasis was given to the features of objects and repeatability [8].
Crack detection and image processing techniques traditionally pursue predefined architectures
that provide the observers with the desired crack detection and classification outputs.
In their review article published in 2017, Mohan and Poobal [9] proposed a famous architecture
for image processing-based crack detection. They suggested that the detection process should
begin with image collection in the architecture. In the next step, the collected images are
preprocessed using gray scaling, smoothing, etc. The key processing algorithms are applied to
the pre-processed images in the third stage. The cracks in the images are then detected using
these processed images. Finally, different attributes such as crack width, length, and depth are
extracted and evaluated in the architecture's final stage. The architecture proposed by them is
given in Fig. 3.
Determination
of research
objectives
Look for
published
research
articles
Selection of
the research
articles based
on methods
and results
Analysis and
synthesis of
the selected
research
artilces.
Representation of
key points and
gaps based on the
literature analysis
Selection of desired
area
Pre-Processing:
>Colour Adjustment
>Noise removal
Image Processing Output

Fig. 3. General structure of crack detection through image processing proposed by Mohan and Poobal [9].
In 2018, Vijayan and Geethalakshmi [10] proposed a quite similar but simplified architecture.
The first phase in their proposed architecture is image collection or data set formation. After that,
preprocessing methods such as smoothing and filtering are applied to the images in the database.
In a single-stage, image processing and crack detection are combined. Processing algorithms,
such as Otsu thresholding, statistical approaches, and thresholding techniques, are used here.
Finally, CNN or Fuzzy-based algorithms are used to classify the detected images. The
summarizations of the steps is given in Fig. 4.
Fig. 4. Architecture of crack detection by image processing proposed by Vijayan et al. [10].
Liu et al. in 2019 [11], suggested a Full CNN based crack detection method using U-Net which is
given in Fig. 5. Being, a deep neural network-based approach, for parameters tuning and hyper-
parameters tuning the main dataset was divided into two parts, the training set and validation set.
The 19 convolutional layered U-Net was trained with 57 input images from the training set. For
the hyper-parameter tuning the rest of the images were used.
Fig. 5. CNN based crack detection using U-net [11].
The marked output images with defined cracks were received at the output layer after input
images were inserted into the trained and tuned U-Net. Adam's optimizer and K-fold cross
Image
Collection
Pre-Processing:
>Gray Scaling
>Smoothing
Image
Processing
Crack
Detection
Attributes
extraction
Database Creation
• Image Accusition
Image Pre-
processing
• Smoothing
• Filtering
Crack Detection
• Otsu
• Statistical Approach
• Thresholding Method
Crack
classification
• CNN
• Fuzzy

validation were used for optimization and validation. As a result, a more efficient algorithm was
developed.
Fig. 6. Input original image (left) and U-Net image (right) [11].
Ren at al. (2019) [12], suggested a crack detection system for tunnels using an improved fully
convolutional neural network. CrackSegNet was the name of the network. The neural network
was built using a modular design that included encoder, decoder path, and 3X3 convolutions,
followed by a 2X2 max pooling layer. The 409 images in the input dataset were augmented to
create a new larger dataset of 919 images. This dataset was split into a training set and a
validation set in a 4:1 ratio for training purposes. The initial RGB images were converted to
grayscale and binary images before being used in the detection process. Finally, they were
subjected to noise reduction before being inserted into the network. The crack segmentation
algorithm returned the images with only cracks being marked in binary form (Fig. 7).
Fig. 7. Work flow of CrackSegNet network to detect cracks [12].

4. Analysis of publication activity dynamics
For this review article, leading peer reviewed journals and conference papers published in
between 2015 to 2020, were surveyed and critically analyzed.
One of the very first papers on accurately evaluating and measuring values from concrete
fractures was published in 2002. The experiment was carried out on concrete blocks [13].
Following that, multiple studies on crack detection using image processing were conducted.
However, there has been a significant advancement in the algorithm and edge detection
technique in the last six years. In recent years, proper edge detection and measurement of crack
characteristics have been a significant concern. As a result, research papers were chosen based on
the methodology and application of their research. Fig. 8 represents the publication frequency of
the articles from 2015 to 2020, which has been analyzed in this review article.
Fig. 8. Frequency of published articles on crack detection in between 2015 - 2020.
Word cloud represents the words which have been given more importance and have been used
repeatedly by various research papers. It is a pictorial representation of the works of various
researchers depending on their word usage in the titles. Fig. 9 represents the word cloud which
has been generated using titles and keywords used by the authors from research articles
published from 2015 to 2020. The larger the size of the words the more they have been repeated
as keywords.
Fig. 9. Word cloud representing the keywords extracted from titles of reviewed articles.
From Fig. 9, it can be emphasized that most of the research work has been done on concrete
structures and pavements where detection of the cracks were given more importance. Various
segmentation and use of classifier can also be observed from the word cloud.
0
2
4
6
8
10
12
2015 2016 2017 2018 2019 2020
Published
articles
Publication Year

5. Literature review
In this review article, literature analysis has been divided into two sections. The first section
provides a more comprehensive assessment of the available review articles on crack
identification using various image processing algorithms. The second section is a comparison of
several research articles that attempted to identify cracks and presented different image
processing algorithms to do so.
5.1. Analysis of published review paper
Five major review articles were published in the last ten years dedicated to crack detection by
image processing. The authors of these review papers examined through existing crack detection
research studies that used various image processing algorithms. Machine learning and deep
learning algorithms were included in some of the approaches. Based on the results of the
reviews, the majority of the studies proposed a common architecture. The papers also included
limitations and future scopes from the analysis. Table 1 summarizes the significant features and
information provided by past review publications to better comprehend their significance in this
study field and assist future researchers.
Table 1
Survey of published review article.
Sl. No. Core Features Ref.
1. (a) Systematic analysis in order to highlight research problems.
(b) 50 research articles were surveyed.
(c) Key characteristics of each methods were determined.
(d) Articles were classified depending on their type of image used.
(e) Common architecture were suggested.
(f) Processing methods, level of accuracy, level of error, as well as dataset-based performance
were reviewed.
[9]
2. (a) Common architecture was suggested
(b) Analysis was done based on DL methods.
(c) 15 research article were surveyed.
(d) Finally they proposed that deep learning can be used to improve the identification of cracks
in surfaces.
[10]
3. (a) Focused on various crack detection techniques both old and new as well as their
technological aspects.
(b) Comparison was made between different methods.
(c) Research was categorized based on algorithm type.
(d) 24 literatures were surveyed.
[14]
4. (a) Knowledge about cracking and its sourced were determined.
(b) 112 papers were surveyed.
(c) Existing and emerging, both types of methods were identified with their advantages and
challenges.
(d) Research articles were categorized based on direct and indirect sensing.
(e) Model-based and model free data analysis were reviewed.
[15]
5. (a) Various crack detection techniques, different methodologies adapted on concrete civil
structures were reviewed.
(b) A common architecture was suggested.
(c) Different crack detection techniques were discussed.
(d) The research articles were categorized based on algorithms.
(e) Challenges and recommendation for future studies were given.
[16]

5.2. Review of published research articles
Review of the published research article has been done on a tabulated format. Articles were sub-
divided based on their investigated crack surfaces. Six different tables has been constructed to
represent those articles.
Table 2
Crack detection on traditional concrete structure surface.
Sl.
No.
Image Type/
Sensor
Image processing
Technique
Algorithms Dataset Key parameters/ Comments Ref.
1. Camera image - FCN
(U-Net)
84 images Precision = 0.90 Recall =
0.91
F1 = 0.90
[11]
2. Camera image - FCN
(VGG 16)
40,000 images Max F1 (%) = 89.3 Average
Precision (AP) (%) = 89.3
[17]
3. Visual Sensor
With laser
beam
- YOLO V3 1800 images.
Trained with: Coco
dataset
Accuracy = 94% Precision =
98%
[18]
4. 4K camera Segmentation and
Multiple Noise
Reduction
Fuzzy
Clustering
50 real concrete
photographs.
Recall = 0.8
Precision = 0.9
Detects width of 0.3 mm or
more.
[19]
5. Camera image Canny Edge
Detection and
width estimation
- Images from the walls
of K-Block, Nirma
University
0.20 mm or less wide cracks
went undetected.
[20]
6. Smartphone
camera
CNN
(Efficient
Net)
Dataset collected by
Smartphone photos
from a suspension
bridge.
Accuracy = 0.9911
Precision = 0.9878
Recall = 0.9945
F1 Score = 0.9912
Accuracy (different dataset) =
0.9737
[21]
7. Camera image FAST, ORB, SIFT,
SURF
- - Av. execution time,
Fast = 461.9 ms
ORB = 329.1 ms SIFT =
1476.5 ms SURF = 488.1 ms
ORB and FAST were
preferred
[22]
8. Camera image Otsu thresholding CNN Open source dataset
with 20,000 cracked
image
Accuracy = 98.25%, 97.18%,
and 96.17% for the first,
second, and third classifiers,
respectively
[23]
9. Camera Image Semantic
Segmentation
Mask
R-CNN
100 & 150 images. Accuracy = 0.9921
Sensitivity =0.7847
Specificity= 0.9933 Precision
= 0.4044 F-measure= 0.4994
[24]
10. Camera Image - CNN 851 pictures from
specimens after
mechanical testing.
Accuracy = 92.27% [25]
11. Camera image - Deep CNN
(ConvNet)
More than 500
pavement pictures.
Precision = 0.8696 Recall =
0.9251
F1 = 0.8965
[26]
12. Camera Image Adaptive Threshold
Method
- Two sets.
First with 3 images &
second with 200
concrete surface
images.
TPR = 94.2% [27]
Legend: FCN Fully Convolutional Network, CNN Convolutional Neural Network, R-CNN Region-based
Convolutional Neural Network, FAST Feature from Accelerated Segment Test, ORB Oriented FAST and rotated
BRIEF, SIFT scale-invariant feature transform, SURF Speeded Up Robust Features, TPR True Positive Rate.

Table 3
Crack detection on flexible pavement surface.
Sl.
No.
Image
Type/
Sensor
Image
processing
Technique
Algorithm Dataset Key parameters/
Comments
Ref.
1. Sports
camera
- CNN
(Dense Net
201)
Two datasets.
CFD and EdmCrack
with 1000 images.
Precision = 91.00% Recall
= 93.22%
F1 = 91.99%
[28]
2. Camera
image
Otsu
thresholding
- Collected RGB image Calculated relative error =
3%
[29]
3. Pave
Vision 3D
system
Ostu
thresholding
- 50 Google images. Specificity = 98.8%
Precision = 77.27%
Accuracy = 97.13% F-
Score =76.09%
[30]
4. Pave
Vision 3D
system
- CrackNet-
V
Images from last 5
years on different
pavements. Image
covers an area of 4 x 2
m2
Precision = 84.3%, Recall
= 90.12%
F-1 = 87.12%
[31]
5. Digital
camera
- CNN 2600 RGB images a
distance of 80 to 100
cm.
Recall = 98.0%, Precision
= 99.4% Accuracy =
99.2%
[32]
6. Camera
image
- YOLO V3 From Highway
Bureau. 3800 images
for training sets and
400 for test sets.
Accuracy 88% [33]
7. Camera
image
Unsupervised
image processing
- Two datasets.
First 55 images from
Google search engine
(keyword “pavement
cracks”).
Second dataset is
annotated road crack
image dataset with 329
images.
Suitable as a pre-
processing step and can
provide rough estimation
of damaged area in an
image.
[34]
8. Camera
image
ROI and saliency
map
- Images from a
highway.
Processing time = 20 fps
Accuracy = 89.33%
[35]
9. Camera
Image
- CNN Collected Pavement
images.
Pavement cracks were
successfully calculated.
[36]
10. CCD
Array
Canny-HBT
filter
- Collected crack
images.
PSNR = 11.15 (db)
Entropy = 6.4054 Errors =
0.3699
FSIM = 0.6602
[37]
11. RGB &
Infrared
Images
Retinex,
Hessian-based
method,
Gabor filter,
Otsu and Median
filter
DBN 920 RGB and infrared
images
Infrared + RGB, Precision
= 0.92
F1 Score = 0.93
Recall = 0.91
RGB,
Precision = 0.90
F1 Score = 0.88 Recall =
0.87
[38]
Legend: PSNR peak signal-to-noise ratio, FSIM feature similarity, ROI region-of-interest, DBN Deep Belief
Network, RGB Red Green Blue.

Table 4
Crack detection on concrete tunnel surface.
Sl.
No.
Image Type/
Sensor
Image
processing
Technique
Comments
Ref.
1. Camera image - Deep FCN
(CrackSegNet)
A total of 409
images from tunnel.
IoU = 38.2% Precision
= 63.85% Recall =
47.46%
F1 = 54.45%
[12]
2. Robotic arm for
capturing
images
- CNN
&
Fuzzy
clustering
Images from
Metsovo motorway
tunnel in Greece.
Accuracy = 0.637
FNR = 0.280
FPR = 0.390
F1= 0.494
[39]
3. Camera Image CEM algorithm - Collected 1,000
pictures.
Accuracy = 91.4% [40]
Legend: IoU Intersection over Union, FNR False Negative Rate, FPR False Positive Rate.
Table 5
Crack detection on concrete bridge surface.
Sl.
No.
Image
Type/
Sensor
Image processing
Technique
Comments
Ref.
1. CCD
camera
image
Otsu threshold
segmentation and
modified Sobel
operator
- Collected
Gray Scale
images.
Precision can reach 0.02 mm [41]
2. Camera
image
- YOLO
v4-FPM
CFAR-10 &
COCO
Recall = 0.978
F1 = 0.979
Precision = 0.00368
[42]
3. CCD
camera
Local adaptive Otsu
and Sobel edge
gradient detection
- Images
collected from
bridges.
Algorithm is feasible in the
real-time automatic detection
of concrete bridge cracks.
[41]
Legend: CCD Charge-Coupled Device.
Table 6
Crack detection on rail tracks.
Sl.
No.
Image Type/
Sensor
Image
processing
Technique
Algorithm Dataset Key parameters /
Comments
Ref.
1. Camera Image Level Set
Method
Fuzzy C
Means
Images collected
from railway tracks.
Entropy was 0.0016 for
high resolution image
and 0.020 was for level
set method respectively.
[43]
2. RAILSCOPE
image
acquisition
system
Adaptive
threshold
method
- From NRC Canada
using a RAILSCOPE
image acquisition
system (IAS).
Computational speed
increased
[44]

Table 7
Crack detection on steel structure surface.
Sl.
No.
Image Type/
Sensor
Image
processing
Technique
Algorithm Dataset Key parameters /
Comments
Ref.
1. Multi-
frequency
EM scanner
- Support
Vector
Machine
Scanned from stainless
steel plates and carbon
fiber-reinforced polymer
(CFRP) plates
Detection rate = 89.7%
Training time = 2.784
Testing time = 2.417
[45]
6. Analysis based on literature review
6.1. Analysis based on the level of accuracy
Based on the literature reviewed, accuracy level-based analysis was done to observe the
performance of the approaches in proper detection of the cracks. Research articles reviewed have
been categorized into four different types of grades based on their accuracy percentages: A (100-
91%), B (90-81%), C (80-71%) and D (70-61%). From literature analysis, it was observed that
on 32 research articles between 2015 - 2020, only eleven paper had justified their accuracy level,
among which eight papers have achieved A-grade level accuracy. It can also be seen that out of
the eight papers which have A-grade accuracy level, six articles had used CNN or CNN-based
YOLO architecture for their crack detection model. However, the lowest accuracy level was
observed for [28], which is about 64%. Accuracy level-based analysis result has been given in
Table 8.
Table 8
Grading of reviewed literature based on accuracy level.
Grade Research articles
A (100-91%) [18,21,23–25,30,32,40]
B (90-81%) [33,35]
C (80-71%) -
D (70-61%) [39]
6.2. Analysis based on algorithms
An algorithm based analysis was performed based on the results of the literature survey, and the
outcome was used to form a pie chart in this section. Observing Fig.10, it can be seen that CNN
(Convolutional Neural Network) algorithm has been used extensively for developing crack
identification models. About 38% of literature has used CNN to develop their detection models,
which is because of its low dependency on preprocessing and easier implementability. Fuzzy C-
means Clustering, Deep FCN and YOLO V3/V4 algorithms were the subsequent most used
algorithms with 14% usage among the papers, where YOLO is also a CNN based object
detection algorithm. However, SVM, DBN, R-CNN and CrackNet-V were used at a low context
of only 5%.

Fig. 10. Usage of different algorithms.
6.3. Analysis based on image processing techniques
From the research articles reviewed, the image processing approaches were extracted to form a
bar chart in order to demonstrate the number of usage. The bar chart highlighted that Otsu
thresholding for image segmentation was the most commonly used image processing approach in
the research articles, with five articles employing it. While adaptive threshold, another image
segmentation approach and the Sobel operator method were only utilized in two of the
publications, other techniques such as Canny edge detection, CEM algorithm, Level set method,
and others were only employed in a single paper.
Fig. 11. Usage of different image processing techniques.
38%
14%
14%
5%
14%
5% 5% 5%
Algorithm Based Analysis CNN
Fuzzy C-means Clustering
YOLO V3/V4
SVM
Deep FCN
R-CNN
CrackNet-V
DBN
0
1
2
3
4
5
6
Image Processing Technique Based Analysis

6.4. Factors affecting the accuracy of crack detection process
(I) Image Quality: Image quality plays a vital role for crack detection in a proper manner. If the
image quality is not up to the mark then the noises in the surface can be detected as cracks. So, a
minimum range of the pixels of image is to be is to be determined in order to carry out the work
properly.
(II) Image Processing Technique: Selection of image processing technique and steps are
important factor to process the image acquired accurately to carry out the further investigation.
In this context, from literature analysis it was seen that Otsu thresholding, adaptive thresholding
and semantic segmentation are better performing image processing techniques.
(III) Selection of Algorithm: Selection of algorithms play a vital role in the accuracy of the
whole process, a suitable algorithm selection results into a better performing model with higher
chances of detection. CNN and CNN based algorithms were seen to have a higher accuracy
compared to other algorithms based on the literature analysis.
(IV) Number of Samples and Their Types: For the evaluation of the developed detection
process, number of sample has been used and their wide range of variety plays a key role to
assess the acceptability of the developed process.
7. Challenges and points to give more concern
1) Most of the research paper mainly focuses on the propagation of the cracks in the longitudinal
direction. But, propagation in the transverse direction sometimes plays a crucial role, especially
when the cracks' width needs to be determined. Therefore, longitudinal and transverse in both
direction estimation of the propagation of the cracks should be done.
2) Estimation of the crack depth is very difficult to predict from sequence of images, especially
for the cracks in open surfaces. So, a thermography based algorithm can be a better option to
develop a process for the estimation of the crack depth.
3) Most of the research has been conducted by developing a system, focusing on a definite type
of structure and cracks. So, an independent system which can quantify, locate and classify
various types of different cracks by a common procedure will be more appropriate in order to use
this method in practical analysis.
4) Resolution of the image plays a vital role for the accuracy and proper result. For camera-based
analysis, there should be a minimum level of resolution below which the detection accuracy falls
below the acceptable range.
8. Proposed approach
Image processing is the process of extracting key parameters from images in order to achieve a
specific goal. An approach for image processing-based crack detection based on the information
collected from reviewed research papers has been proposed in this survey study given in Fig. 12.

Fig. 12. Proposed approach for crack detection based on the results of the literature analyzed
The proposed approach is divided into 5 steps:
(1)Image collection/ dataset creation: The first step in the approach is to gather crack images
and create a dataset.
(2)Pre-processing: The next step of the approach includes pre-processing of the images with
smoothing, gray scaling, and noise reduction.
(3)Segmentation: This step includes segmentation of the images using Otsu thresholding for
inserting into the detection process.
(4)Crack detection: In this stage, detection algorithm such as Convolutional Neural Networks
can be used to detect either crack or non-crack images.
(5)Detection of crack attributes: The final step involves using classifiers again for the purpose
of detection of crack length, width and depth.
9. Conclusion
Crack identification through image processing is a novel technique that reduces the time and cost
to identify the cracks in the structure. In this review paper, a number of published articles
depending on their experimental structure, steps followed, and outcomes, have been reviewed to
make a summary and to justify the accuracy of image processing-based crack detection. After the
literature review, based on the key information gathered from the survey, analysis was made to
point out accuracy level, usage of the algorithm, and the key factors that affect this technique's
accuracy. Based on the surveyed research articles, challenges and the critical points needed to
give more concern have been figured out to help the researchers for developing a crack

identification system that will be unique and accurate. It was observed that camera-based image
processing has a great interest among the researchers due to its lower cost and multiple
approaches. But the resolution of the images plays a vital role in this technique. The highest
accuracy level was also observed in camera-based analysis, but steps for identification of the
crack depth were missing in most of the papers. Moreover, the accuracy of the approaches to
assess the developed method was not given in most of the research articles. Depending on the
survey, a new image processing structure to detect cracks with all its parameters was proposed.
The proposed structure was developed by synthesis of various approaches and based on their
accuracy results. But, for developing a unique system, lots of practical implementations are
needed. So, it can be concluded that image processing-based crack detection can be an excellent
alternative to reduce the difficulties of human-based time-consuming approaches. But a general
structure that will be fully applicable to any structure surface is needed.
In the future, we aim to develop a crack detection model based on the proposed approach given
in this research study, which will detect the crack attributes with higher accuracy.
Acknowledgement
The authors declared no conflict of interest. The authors would like to thank Bangladesh
University of Professionals and Chittagong University of Engineering and Technology for
supporting this research work.
References
[1] Budiansky B, O’connell RJ. Elastic moduli of a cracked solid. Int J Solids Struct 1976;12:81–97.
doi:10.1016/0020-7683(76)90044-5.
[2] Torok MM, Golparvar-Fard M, Kochersberger KB. Image-Based Automated 3D Crack Detection
for Post-disaster Building Assessment. J Comput Civ Eng 2014;28. doi:10.1061/(ASCE)CP.1943-
5487.0000334.
[3] Jahangir H, Esfahani MR. Structural Damage Identification Based on Modal Data and Wavelet
Analysis. 3rd Natl. Conf. Earthq. Struct., 2012.
[4] Kim H, Ahn E, Cho S, Shin M, Sim S-H. Comparative analysis of image binarization methods for
crack identification in concrete structures. Cem Concr Res 2017;99:53–61.
doi:10.1016/j.cemconres.2017.04.018.
[5] Oliveira H, Correia PL. Automatic Road Crack Detection and Characterization. IEEE Trans Intell
Transp Syst 2013;14:155–68. doi:10.1109/TITS.2012.2208630.
[6] Fujita Y, Hamamoto Y. A robust automatic crack detection method from noisy concrete surfaces.
Mach Vis Appl 2011;22:245–54. doi:10.1007/s00138-009-0244-5.
[7] Garber D, Shahrokhinasab E. Performance Comparison of In-Service, Full-Depth Precast Concrete
Deck Panels to Cast-in-Place Decks. Accelerated Bridge Construction University Transportation
Center (ABC-UTC); 2019.
[8] Niemeier W, Riedel B, Fraser C, Neuss H, Stratmann R, Ziem E. New digital crack monitoring
system for measuring and documentation of width of cracks in concrete structures. Proc. 13th FIG
Symp. Deform. Meas. Anal. 14th IAG Symp. Geod. Geotech. Struct. Eng. Lisbon, 2008, p. 12–5.
[9] Mohan A, Poobal S. Crack detection using image processing: A critical review and analysis.
Alexandria Eng J 2018;57:787–98. doi:10.1016/j.aej.2017.01.020.
[10] Geethalakshmi SN. A survey on crack detection using image processing techniques and deep

learning algorithms. Int J Pure Appl Math 2018;118:215–20.
[11] Liu Z, Cao Y, Wang Y, Wang W. Computer vision-based concrete crack detection using U-net fully
convolutional networks. Autom Constr 2019;104:129–39. doi:10.1016/j.autcon.2019.04.005.
[12] Ren Y, Huang J, Hong Z, Lu W, Yin J, Zou L, et al. Image-based concrete crack detection in tunnels
using deep fully convolutional networks. Constr Build Mater 2020;234:117367.
doi:10.1016/j.conbuildmat.2019.117367.
[13] Ito A, Aoki Y, Hashimoto S. Accurate extraction and measurement of fine cracks from concrete
block surface image. IEEE 2002 28th Annu. Conf. Ind. Electron. Soc. IECON 02, vol. 3, IEEE;
n.d., p. 2202–7. doi:10.1109/IECON.2002.1185314.
[14] Bhat S, Naik S, Gaonkar M, Sawant P, Aswale S, Shetgaonkar P. A Survey On Road Crack
Detection Techniques. 2020 Int. Conf. Emerg. Trends Inf. Technol. Eng., IEEE; 2020, p. 1–6.
doi:10.1109/ic-ETITE47903.2020.67.
[15] Yao Y, Tung S-TE, Glisic B. Crack detection and characterization techniques-An overview. Struct
Control Heal Monit 2014;21:1387–413. doi:10.1002/stc.1655.
[16] Anitha MJ, Hemalatha R, Radha S. A Survey on Crack Detection Algorithms for Concrete
Structures, 2021, p. 639–54. doi:10.1007/978-981-15-5029-4_53.
[17] Dung CV, Anh LD. Autonomous concrete crack detection using deep fully convolutional neural
network. Autom Constr 2019;99:52–8. doi:10.1016/j.autcon.2018.11.028.
[18] Park SE, Eem S-H, Jeon H. Concrete crack detection and quantification using deep learning and
structured light. Constr Build Mater 2020;252:119096. doi:10.1016/j.conbuildmat.2020.119096.
[19] Noh Y, Koo D, Kang Y-M, Park D, Lee D. Automatic crack detection on concrete images using
segmentation via fuzzy C-means clustering. 2017 Int. Conf. Appl. Syst. Innov., IEEE; 2017, p.
877–80. doi:10.1109/ICASI.2017.7988574.
[20] Vora J, Patel M, Tanwar S, Tyagi S. Image Processing Based Analysis of Cracks on Vertical Walls.
2018 3rd Int. Conf. Internet Things Smart Innov. Usages, IEEE; 2018, p. 1–5. doi:10.1109/IoT-
SIU.2018.8519926.
[21] Su C, Wang W. Concrete Cracks Detection Using Convolutional NeuralNetwork Based on Transfer
Learning. Math Probl Eng 2020;2020:1–10. doi:10.1155/2020/7240129.
[22] Kong X, Zhang Z, Meng L, Tomiyama H. Machine Learning Based Features Matching for Fatigue
Crack Detection. Procedia Comput Sci 2020;174:101–5. doi:10.1016/j.procs.2020.06.063.
[23] Flah M, Suleiman AR, Nehdi ML. Classification and quantification of cracks in concrete structures
using deep learning image-based techniques. Cem Concr Compos 2020;114:103781.
doi:10.1016/j.cemconcomp.2020.103781.
[24] Yamane T, Chun P. Crack Detection from a Concrete Surface Image Based on Semantic
Segmentation Using Deep Learning. J Adv Concr Technol 2020;18:493–504.
doi:10.3151/jact.18.493.
[25] Silva WRL da, Lucena DS de. Concrete Cracks Detection Based on Deep Learning Image
Classification. Proceedings 2018;2:489. doi:10.3390/ICEM18-05387.
[26] Zhang L, Yang F, Daniel Zhang Y, Zhu YJ. Road crack detection using deep convolutional neural
network. 2016 IEEE Int. Conf. Image Process., IEEE; 2016, p. 3708–12.
doi:10.1109/ICIP.2016.7533052.
[27] Liu X, Ai Y, Scherer S. Robust image-based crack detection in concrete structure using multi-scale
enhancement and visual features. 2017 IEEE Int. Conf. Image Process., IEEE; 2017, p. 2304–8.
doi:10.1109/ICIP.2017.8296693.
[28] Mei Q, Gül M, Azim MR. Densely connected deep neural network considering connectivity of
pixels for automatic crack detection. Autom Constr 2020;110:103018.
doi:10.1016/j.autcon.2019.103018.
[29] Shao C, Chen Y, Xu F, Wang S. A Kind of Pavement Crack Detection Method Based on Digital

Image Processing. 2019 IEEE 4th Adv. Inf. Technol. Electron. Autom. Control Conf., IEEE; 2019,
p. 397–401. doi:10.1109/IAEAC47372.2019.8997810.
[30] Akagic A, Buza E, Omanovic S, Karabegovic A. Pavement crack detection using Otsu thresholding
for image segmentation. 2018 41st Int. Conv. Inf. Commun. Technol. Electron. Microelectron.,
IEEE; 2018, p. 1092–7. doi:10.23919/MIPRO.2018.8400199.
[31] Fei Y, Wang KCP, Zhang A, Chen C, Li JQ, Liu Y, et al. Pixel-Level Cracking Detection on 3D
Asphalt Pavement Images Through Deep-Learning- Based CrackNet-V. IEEE Trans Intell Transp
Syst 2020;21:273–84. doi:10.1109/TITS.2019.2891167.
[32] Yusof NAM, Osman MK, Noor MHM, Ibrahim A, Tahir NM, Yusof NM. Crack Detection and
Classification in Asphalt Pavement Images using Deep Convolution Neural Network. 2018 8th
IEEE Int. Conf. Control Syst. Comput. Eng., IEEE; 2018, p. 227–32.
doi:10.1109/ICCSCE.2018.8685007.
[33] Nie M, Wang C. Pavement Crack Detection based on yolo v3. 2019 2nd Int. Conf. Saf. Prod.
Informatiz., IEEE; 2019, p. 327–30. doi:10.1109/IICSPI48186.2019.9095956.
[34] Buza E, Akagic A, Besic I. Image-Based Crack Detection Using Sub-image Technique. 2019 11th
Int. Conf. Electr. Electron. Eng., IEEE; 2019, p. 614–8. doi:10.23919/ELECO47770.2019.8990615.
[35] Kang S-M, Chun C-J, Shim S-B, Ryu S-K, Baek J-D. Real Time Image Processing System for
Detecting Infrastructure Damage: Crack. 2019 IEEE Int. Conf. Consum. Electron., IEEE; 2019, p.
1–3. doi:10.1109/ICCE.2019.8661830.
[36] Qin J, Zhu Q, Li L, Dong L. Pavement Cracks Automated Processing Based On Image Detection
with Neutral Network. 2020 12th Int. Conf. Intell. Human-Machine Syst. Cybern., IEEE; 2020, p.
103–6. doi:10.1109/IHMSC49165.2020.00031.
[37] Wu G, Sun X, Zhou L, Zhang H, Pu J. Research on crack detection algorithm of asphalt pavement.
2015 IEEE Int. Conf. Inf. Autom., IEEE; 2015, p. 647–52. doi:10.1109/ICInfA.2015.7279366.
[38] Jo J, Jadidi Z. A high precision crack classification system using multi-layered image processing
and deep belief learning. Struct Infrastruct Eng 2020;16:297–305.
doi:10.1080/15732479.2019.1655068.
[39] Doulamis A, Doulamis N, Protopapadakis E, Voulodimos A. Combined Convolutional Neural
Networks and Fuzzy Spectral Clustering for Real Time Crack Detection in Tunnels. 2018 25th
IEEE Int. Conf. Image Process., IEEE; 2018, p. 4153–7. doi:10.1109/ICIP.2018.8451758.
[40] Niu B, Wu H, Meng Y. Application of CEM Algorithm in the Field of Tunnel Crack Identification.
2020 IEEE 5th Int. Conf. Image, Vis. Comput., IEEE; 2020, p. 232–6.
[41] Wang Y, Zhang JY, Liu JX, Zhang Y, Chen ZP, Li CG, et al. Research on Crack Detection
Algorithm of the Concrete Bridge Based on Image Processing. Procedia Comput Sci
2019;154:610–6. doi:10.1016/j.procs.2019.06.096.
[42] Yu Z, Shen Y, Shen C. A real-time detection approach for bridge cracks based on YOLOv4-FPM.
Autom Constr 2021;122:103514. doi:10.1016/j.autcon.2020.103514.
[43] Chaudhari CV, Gupta RK, Feagade SA. A Novel Approach of Crack Detection in Railway Track
using Fuzzy C Means and Level Set Method. 2nd Int. Conf. Data, Eng. Appl., IEEE; 2020, p. 1–7.
doi:10.1109/IDEA49133.2020.9170669.
[44] Sambo B, Bevan A, Pislaru C. A novel application of image processing for the detection of rail
surface RCF damage and incorporation in a crack growth model. Int. Conf. Railw. Eng. (ICRE
2016), Institution of Engineering and Technology; 2016, p. 12 (9 .)-12 (9 .).
doi:10.1049/cp.2016.0521.
[45] Yin L, Wu J, Ye B, Avila JS, Xu H, How KY, et al. Imaging and detection of cracks based on a
multi-frequency electromagnetic scanning instrument and SVM. 2017 IEEE Int. Conf. Imaging
Syst. Tech., IEEE; 2017, p. 1–5. doi:10.1109/IST.2017.8261512.

A Comparative Review of Image Processing Based Crack Detection Techniques on Civil Engineering Structures

More Related Content

Similar to A Comparative Review of Image Processing Based Crack Detection Techniques on Civil Engineering Structures (20)

More from Journal of Soft Computing in Civil Engineering (20)

Recently uploaded (20)

A Comparative Review of Image Processing Based Crack Detection Techniques on Civil Engineering Structures