A Review of various Surface Defect Detection Techniques

Chetan V. Chaudhari
Research Scholar, RKDF Institute of Science & Technology, SRK University, Bhopal
Department of Computer Engineering, J. T. Mahajan College of Engineering Faizpur

Abstract

In this paper, we review various surface defect detection techniques using image
processing. The aim is to review the techniques for some methods of surface defect
detection techniques. These methods are able to differentiate the features extracted
from normal and defective regions. It is impossible to cover all the detail aspects
of each scheme here due to vast field. This paper mainly focuses on a particular but
important part of some methods or scheme of surface defect detection. Other topics
related to surface inspection such as imaging system, data acquisition and
correction are out of the scope of this paper.

Defect detection from images is becoming increasingly significant in a variety

of applications since quality inspection is an important aspect of modern industrial
manufacturing. Defect detection methods are mainly categorized in four
approaches statistical, structural, filter based methods, and model based
approaches.

Key Words: Surface Inspection, Defect Detection.

1 Introduction

Quality inspection is an important aspect of modern industrial manufacturing. The target

to be achieved through this project was primarily aimed at detecting the surface defects
belonging to different classes. This was achieved through taking the images from the camera,
carrying out defect detection on these images and classifying them. We present a method to
automatically detect and localize defects occurring on the surface. Defect regions are
segmented from background images using their distinguishing texture characteristics.
As an example a visual inspection for texture and color characteristics has application on a
various surfaces e.g. steel, wafer, ceramics, wood, fruits and aircraft surfaces, and is highly
desirable by industry in order to repeat the process of manual inspection. Example in ceramic
tile production, chromate-texture characteristics of the final aspect can be affected by a variety
of extern facts that are hard to control, such as color pigments, humidity, and temperature.
Therefore, online monitoring with feedback control on whole production line becomes
desirable.
We review surface defect detection as the texture feature extraction and analysis in four
categories, statistical, structural, filter based, and model based approaches. We also reviewed
relative studies with this.
2 Brief Review of Methods

Defect detection methods are mainly categorized in statistical approach, spectral

approach and model based approach. These approaches are briefly described as below:
2.1 Statistical Approach
Methods in this approach mainly focus on statistical behavior of different regions of
image. Statistical approach describes spatial distribution of gray level by two
representations auto-correlation function and co-occurrence matrix.
Auto-correlation Function
Auto-correlation function measures spatial frequency of image and it gives maxima of that
frequency at different locations according to the length of repetitive primitive on image.
These maxima will be constant for the primitive that has been perfect throughout the
image and different for the primitives that are changed and imperfect in replication. As a
result those primitives can be considered as defective this method is mostly used for
regular patterned images. It measures regularity and coarseness of pattern. But this
method has limitation. It needs reference frame of tonal primitive to carry out analysis of
texture.
Co-occurrence Matrix
Co-occurrence matrix is the most widely used method for texture classification. It uses 2D
matrices to accumulate various texture features of images such as energy, contrast,
entropy, correlation, homogeneity etc. These texture features are characterized as second
order statistic which is the measure of spatial dependence of gray values for specific
 distance. This method has some limitations. The size of co-occurrence matrix is
important. So number of gray values must be reduced to meet the memory requirements.
If the texture features are constructed using large sized primitive than this methods shows
poor performance.
Mathematical Morphology
Mathematical morphology takes out useful components from image for the description
and representation of regional shape. In this method operations such as erosion, dilation,
opening and closing are performed on image using structuring element. The benefit of
this method is that it gives response to various defect size and shapes. It is also better for
segmentation. It is mostly suitable for unidirectional textures.
Edge Detection
Edge detection techniques are also very effective in detection of defects. The distribution
of number of edges is the important feature in texture images. In an image point, line and
edge defects can be represented using number of gray level transition in an image. These
features can be used to detect defects. But this method has also some drawbacks. This
approach is only suitable to plain weave fabric images. With these method defects nearby
edges are hard to detect.
2.2 Spectral Approach
Methods in these approaches are applicable when texture images are composed of
reoccurrence of some basic primitive with acceptance of specific rules of displacement.
Therefore, spectral approaches are not suitable for the images with random texture
features.
Fourier Transform
Fourier transform can be derived from Fourier series. While spatial domain is sensitive to
noise and difficult to detect defect, Fourier transform utilizes frequency domain to detect
defects. This transform has the properties of noise immunity, optimal characterization of
periodic features and translation invariance. Fourier transform can be categorized in two
categories: Discrete Fourier transform and Optical Fourier transform. The DFT based
approaches are ineffective for the images in which frequency component of defects
 appear in image are highly mixed with each other in frequency domain. In OFT, defect
detection of fabric image is very easy and fast because it is obtained in optical domain by
using lenses and spatial filter.
Wavelet Transform
Wavelet transform is another spectral approach for defect detection. Wavelet represents
decomposition of multi-resolution signal. Fourier transforms are sinusoidal whereas
wavelet transform are small waves of varying frequency and some specific duration
called wavelets. Wavelet transform provides more local support from vertical, horizontal
and diagonal direction for any inputted image. The multi-scale wavelet representation has
the property of shift invariance and it can detect defects by examining image at different
scales.
Gabor transform
The general form of Gabor function is in a nonorthogonal basis set. Gabor filter provides
optimal joint localization in both spatial and spatial-frequency domain. The texture
features that represent frequency content in local region in spatial domain can be
extracted by localized spatial filtering. Gabor filters provide this type of filtering. The
implementation of Gabor filter is categorized in two ways: 1) Filter bank consisting group
of filters with predetermined parameters in frequency and orientation to adequately cover
frequency plane.
2) Implementation of optimal filters with the use of few filters but correct choice for that
filters is hard and crucial.
2.3 Filtering approach
Filtering technique is used in many applications to filter out image for smoothening of
image by suppressing high frequency or for enhancing image by suppressing low
frequency. Filtering is performed between image neighborhood and filtering mask. In
Filtering approach there are three kinds of methods:
a) Spatial filtering which is applied directly on pixels
b) Frequency filtering which is based on Fourier transforms.
c) Joint spatial-frequency filtering.
In spatial domain, images are filtered out by using gradient filters to extract dots, lines and
edges. In this approach first Sobel, Canny, Robert, Laplacian, Daubechies and Law filters
are used to measure edge density. Many other methods use frequency domain filtering
approach in the case when there are no straightforward kernels can be found. In this
approach image is first transformed to Fourier domain, multiplied with filter function and
then again re-transformed to spatial domain. Example of frequency domain filters are
Ring filter and Wedge filters.
2.4 Model based Approach
Texture is usually considered as a complex pictorial pattern. Any random field in mage
can be defined by stochastic model which is modeled by simple function of an array of
random variables. This modeling based approach has advantage that it can produced
texture that can match the observed texture. Model based approaches are mostly suitable
to fabric images with stochastic surface variations or for randomly textured fabrics for
which the statistical and spectral approaches have not yet shown good results. Model
based methods include autoregressive model, fractal model, markov random field model
and the Texem model.
The comparison of the different methods used for the defects detection in images are as
below
1. Edge Detection:
It identifies defects accurately and effectively but not working with defects other than
crack and holes.
2. Thresholding:
It is a good method due to automatically select the threshold value but it required clear
peek and valleys of threshold.
3. Morphological Operation:
It is a good for smooth images with less light disturbance but this method is very
sensitive to defect shape size.
4. Co-occurrence Matrix:
 It is good for images which has unchanging glow but only works on unchanging
domain conditions as well as it required more memory and computations.
5. Wavelet Transform:
It identifies defect based on line like diagonal, horizontal and vertical line defects but
failed in color variance and edges in images.

3 Conclusions

These review of various surface defect detection techniques using image processing gives us
some accurate state-of-the-art and possible trend to application area for manufacturing
industry. This paper gives surface defect detection methods with advantages and
disadvantages along with some comparative studies.

As with our example of texture analysis, statistical and filter based approaches are preferred.
For manufacturing industry real time performance is highly desirable.

We need of some standard datasets with experimental protocols to study comparative analysis
of various surface defect detection techniques.

4 References

[1] Xianghua Xie, “A Review of Recent Advances in Surface Defect Detection using
Texture analysis Techniques” Electronic Letters on Computer Vision and Image
Analysis 7(3):1-22, 2008
[2] F. Ade. Characterization of texture by ‘eigenfilter’. Signal Processing, 5(5):451–457,
1983.

[3] F. Ade, N. Lins, and M. Unser. Comparison of various filter sets for defect detection
in textiles. In International Conference on Pattern Recognition, volume 1, pages 428–
431, 1984.

[4] Automated fabric defect detection—A review, Henry Y.T. Ngan, Grantham K.H. Pang
, Nelson H.C. Yung, Image and Vision Computing 29 (2011) 442–458.
[5] Computer-Vision-Based Fabric Defect Detection: A Survey Ajay Kumar, Member,
IEEE, IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO.
1, JANUARY 2008.

[6] An Algorithm to Detect and Identify Defects of Industrial Pipes Using Image
Processing, Md. Ashraful Alam, M MNaushad Ali, Musaddeque Anwar AI-Abedin
Syed, Nawaj Sorif, Md. Abdur Rahaman, 978-1-4799- 6399-7114/$31.00 ©20 14
IEEE.

[7] Fabric Defect Detection Algorithm Using Morphological Processing and DCT,
Mahmoud Abdel Aziz, Ali S. Haggag and Mohammed S. Sayed, Dept. of Electronics
and

[8] Communications Engineering, Zagazig University, Zagazig, Egypt, 978-1-4673-2821-

0/13/$31.00 ©2013

[9] Metal Surface Defect Detection using Iterative thresholding Technique,

M.Senthikumar, Dr.V.Palanisamy, Dr.J.Jaya, 2nd International Conference on Current
Trends in Engineering and Technology, ICCTET’14, © IEEE 2014.

[10] Fabric defect detection based on GLCM and Gabor filter: A comparison, Jagdish Lal
Raheja, Sunil Kumar, Ankit Chaudhary, Optik 124 (2013) 6469– 6474.

[11] Fabric Defect Detection using Wavelet Filter, Mr. Vaibhav V. Karlekar, Prof.
M.S.Biradar & Mr. K.B.Bhangale, 2015 International Conference on Computing
Communication Control and Automation, 978-1-4799-6892-3/15 $31.00 © 2015
IEEE.

[12] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, Z. Wojna, "Rethinking the inception

architecture for computer vision", 2015.

[13] Smriti Bhandari , S.M. Deshpande, “ A Simple Approach to Surface Defect Detection.
2008 IEEE Region 10 Colloquium and the Third International Conference on Industrial
and Information Systems, Kharagpur, INDIA December 8 -10, 2008.