2019 13th International Conference on Mathematics, Actuarial Science, Computer Science and Statistics (MACS)

Automatic Fabric Fault Detection Using Image

Processing
Engr.Anum Khowaja Engr.Dinar Nadir
Department Of Information Technology Department Of Electronics Engineering
Barret Hodgson University Dawood University Of Engineering And Technology
Karachi, Pakistan Karachi, Pakistan
an_alwani@hotmail.com dinarnadir@hotmail.com

Abstract— This paper provides an overview of automatic fabric • Weaving defects - The defects that occur during
fault detection approaches that have been developed in recent weaving.
years. Fabric fault detection is very popular topic of automation
moreover quality control is one of the important features in
textile industry. The performance of the projected idea is
evaluated by using different techniques of patterned fabric
images with different types of common fabric defects. Moreover
detection methods were also evaluated in real time using a model
automation specification system. This research paper will be Fig. 1. Yarn and weaving defects
useful for both researchers and practitioners in the field of image
processing and computer vision to understand the uniqueness of
the different defect detection methods. The recognition receives a II. LITERATURE REVIEW
digital fabric image from the image acquisition device and Over the last two years, fabric defect detection by means
transforms it to a binary image using the restoration and of image processing has received considerable attention, and
threshold methods. This research presents a technique that several methods have been suggested in the literature. Cho et
decreases physical exertion. This image processing method was
al. [1] proposed an inspection method that works at the line
performed using "MATLAB 7.10". Therefore, this study uses a
level to detect defects in uniform textures.
textile fault detector with a systematic vision approach for image
processing. Tajeripour et al. [2] used local binary compression (LBP)
to detect defects in fabric. In this work, a training phase was
first performed by applying the B-slide to the non-disabled
Keywords—Rgb; threshold; image processing; filtering tissue image. The disadvantages are found in the new image
using the correct level. However, the convenience of these
I. INTRODUCTION methods is restricted to uniform (non-patterned) textures.
Automatic inspection of stained cloth has been an Chan and Pang used Fourier analysis to detect defects in the
admirable research topic in manufacturing and quality control gray images [9].
for over twenty years. It aims to identify and paint defects on Kumar and Pang [3] used a variety of filters that controlled
the surface of stained fabric at the stage of production. In the and trained to detect defects in textured materials. However,
past it was mainly achieved through the supervision of skilled the Gore technique usually requires a lot of computation,
workers, but there were some disadvantages such as high error while the integration of different channels is still an open
rates due to human fatigue, high labor costs and slow question.
verification speed. Automatic vision monitoring improves the Recently, Nan et al. used wavelet transforms to detect gaps
effectiveness of such controls and provides satisfactory in patterned textures. This method can detect defects for
detection accuracy for quality control in the textile industry. different types of textures. However, it does use a sensitivity
To overcome these hindrances, a system of image processing level that is manually locked [8].
is employed using MATLAB 7.10. A smooth fabric is taken as
a sample and the defects will be observed on the basis on More recently, Cézar et al. [7] used independent finite
roughness and tear. To get a proper result we use 512 X 512 element analysis (IRA) to detect block-level errors in texture
images. This method works best with uniform gray level
camera in the acquisition of data. In the textile industry, some
texture. However, this does not summarize the model with
common defects occur are discussed below:
good texture and many disadvantages are not found in this
method.
• Yarn defects - The defects arising from the rotating or
winding stage. In Tilocca and Antonio [4] the researchers introduced a
new method for the automatic fabric testing based on optical
system of acquisition using an artificial neural network (ANN)

978-1-7281-4956-1/19/$31.00 ©2019 IEEE

Authorized licensed use limited to: University of Exeter. Downloaded on June 15,2020 at 14:56:49 UTC from IEEE Xplore. Restrictions apply.
to study the collected data. They also proved that a direct processing methods are required in the image. The noise
automatic method with textile controls can achieve good should be removed from the image by the noise reduction
results using optical technology and powerful data processing technique [9].
research capabilities. The large amount of data that can be
collected with the investigation in a relatively short period of B. Gray Image Conversion
time seems sufficient to obtain a rapid and accurate Converting an image to a grayscale is important because
classification of pre-qualified ANNs without any other data continuous processing of the system should be performed only
exchange [6]. on gray images. A RGB image is a three layered image having
New methods for categorization of fabric defects have red scale, blue scale and green scale. Similarly a gray scale
been designed and implemented to meet the needs of image is a single layered image. A gray digital image is an
manufacturers. An analysis was conducted in the industrial image where the value of each pixel is a single sample. The
environment to highlight the most costly and destructive Image intensity values range from 0 to 255.To convert RGB
defects occur in fabric. The most repeated defects are missing image to gray scale image a function called “RGB2GRAY()”
weft or wrap threads, oil stains and holes [3]. is available in MATLAB [10].
Jackson [5] came up with a concept for automatic fabric
C. Noise Removal And Filtering
detection through image processing and artificial neural
networks. Mostly noise produces at the time of capturing or image
transmission. The noise means that the pixels in the image
show different intensities from the true pixel values obtained
III. METHODOLOGY from the images. The noise removal algorithm is the process
Digital analysis of 2D fabric images is created on the basis of removing or reducing the noise in the transformed image.
of receiving images from a system. The system must be able
to accept a defective fabric image and then convert it to a D. Threshold And Histogram Equalization
grayscale image. The appropriate noise exclusion of the image
In the perspective of image histogram processing, images
must be applied and transformed to the equivalent binary
are usually referred to as histograms of pixel intensity values.
image. The result must be in the form of a histogram.
This histogram is a graph that gives you an idea about the
Furthermore to the histogram properties, a well-defined
number of pixels in an image at each value of the different
threshold function is also measured for the results. The whole
intensities contained in that image. In simple terms, it signifies
system can be monitored as shown in Fig.2. A detailed
the number of pixels for separately intensity value measured.
explanation and task of each block is summarized as follows.
Histogram equalization is used to enhance the contrast.
Contrast is the variation in color that makes an object
distinguishable from other objects in the same field of view
moreover it is a computerized image processing method
utilizes to get better image contrast. It is achieved by
distributing the most common intensity values effectively.
Extend the image intensity range. This method typically
enhances the contrast of the global image when its available
data is represented by close contrast values. This permits the
local low-level regions to achieve high contrast.

Fig. 2. The flow system of image processing

A. Capture Image
This is basically an image recognition unit. In this unit the
faulty image is captured by a digital camera having 320 X 420
pixels. Digital images can be generated by a variety of input
devices and methods such as digital cameras, scanners and
coordinate measuring machines. The acquired image probably
Fig. 3. Flow chart of histogram equalization
cover noise signals, if it holds noise signals, then some prior

Authorized licensed use limited to: University of Exeter. Downloaded on June 15,2020 at 14:56:49 UTC from IEEE Xplore. Restrictions apply.
Table 1. Histogram equalization table
IV. RESULTS
Gray nK PDF CDF (L- ROUND The following test image was used to see the defects in the
fabric. The picture has been presented to histogram
Level 1)*CDF OFF equalization calculation for threshold, where the uneven
VALUES weaving is identified as spots.
0 780 0.119 0.109 1.3 1
1 1093 0.225 0.404 3.0 3 I = imread('sample01.jpg');
2 852 0.261 0.650 4.5 5 img_mat = rgb2gray(I);

3 655 0.106 0.810 5.6 6 [m,n] = size(img_mat);

4 320 0.018 0.809 6.2 6 bitlevel = 8;
5 244 0.016 0.905 6.6 7
bit_comb = (2^bitlevel);
6 111 0.003 0.908 6.8 7
7 92 0.012 1 7 7 hist_data = zeros(1,bit_comb);

4092 %Histogram
for i = 1:m
for j = 1:n
val = img_mat(i,j)+1;
Table 2. CDF calculation count = hist_data(val) + 1;
hist_data(val) = count;
end
GRAY LEVEL CDF CDF* LEVEL-1 end
VALUE
0 0.110 0
1 0.21 3 %streching
max_pix_vec = max(img_mat(1:m,1:n));
2 0.54 3 min_pix_vec = min(img_mat(1:m,1:n));
3 0.60 4
4 0.77 7 max_pix = max(max_pix_vec);
min_pix = min(min_pix_vec);
5 0.86 6
6 0.92 6 for i= 1:m
for j = 1:n
streched_img(i,j) = (img_mat(i,j) -
min_pix)*((bit_comb - 1)/(max_pix - min_pix));
The purpose of leveling or thresholding is to remove those end
pixels from a number of images that represent objects (such as end
a graphical map). This method can be determined by looking
at the image intensity histogram. There are two fundamental hist_data_new = zeros(1, bit_comb);
types of thresholding discussed in this research work for i = 1:m
for j = 1:n
val = streched_img(i,j)+1;
• Global thersholding count = hist_data_new(val)+1;
• Local thresholding hist_data_new(val) = count;
end
end
Global level settings have an intensity value (threshold), so
every unit with an intensity value below the threshold belongs max_pix_vec_new = max(streched_img(1:m,1:n));
to one stage, the remainder belongs to one. The global level is min_pix_vec_new = min(streched_img(1:m,1:n));
as good as the degree of intensity separation between the two
max_pix_new = max(max_pix_vec_new);
peaks in the image. Basic-level adapters are only used to min_pix_new = min(min_pix_vec_new);
convert pixels from grayscale to black and white pixels.
Normally, pixel value 0 is white and 255 is black, with old_con = max_pix - min_pix;
numbers 1 through 254 signifying different levels of gray. new_con = max_pix_new - min_pix_new;
Unlike the global thresholding technique, local adaptive figure(1);
thresholding selects altered threshold values for each pixel in subplot(2,2,1)
the image based on an analysis of its neighboring pixels. This stem(0:bit_comb-1, hist_data)
title('old image')
is to allow images with fluctuating contrast levels where a
global thresholding technique will not work adequately. subplot(2,2,2)
Different forms of adaptive thresholding algorithms have been imshow(img_mat)
reported in the image processing literature. title('old image')

Authorized licensed use limited to: University of Exeter. Downloaded on June 15,2020 at 14:56:49 UTC from IEEE Xplore. Restrictions apply.
subplot(2,2,3)
stem(0:bit_comb-1, hist_data_new)
title('image histogram')

subplot(2,2,4)
imshow(streched_img)
title('new image')

Fig. 6. Contrast stretched image

V. CONCLUSION
It is easy to see defects in an image using this method.
Manually inspecting textile quality usually goes through the
Fig. 4. Histogram output inspection of the human eye. Monitoring one's vision is a
tedious task that involves observing, paying attention, and
attempting to accurately detect the occurrence of failure. The
system can detect tissue defects with greater accuracy and
efficiency. In the textile industry, we can detect the damage in
the real picture and we can repair the damage with the help of
advanced management systems. In the textile industry,
damage can be detected by wireless consent. We used
MATLAB in this file, but new software such as SDL, Virtual
LB and Computer Vision may be used in the future. The
method which is used in this research work categorizes 90%
of faults in the cloth. Improvement and enhancement of visual
system performance can generally be done with the proposed
algorithm to detect common defects in normal textures. In
addition, it can be applied to models combining different light
elements without much adaptation. The flexibility of this
method is verified not only by the availability of different
common textures, but also of the texture, the method allows
for the detection of different defects.

REFERENCES
Fig. 5. RGB to gray contrast stretched

[1] Cho, T.S., “Image RestorationBy Matching Gradient Distributions”,

IOSR Journal of Engineering, Vol. 2, No.4, pp. 582-584, Apr. 2012.
[2] Tajaripour.S,“A Review of Recent Advances in Surface Defect
Detection usingTexture analysis Techniques”, Electronic Letters on
Computer Vision and Image Analysis, Vol.7, No.3, pp.1-22, 2008.
[3] Kumar and Pang, R.A. Campbell and R.J. Harwood, “Automated
inspection of carpets”, in Proc. SP IE, Vol. 2345, pp. 180-191, 1995.
[4] P.M. Jackson, S.R. Kolhe and P.M. Patil “A Review of Automatic
Fabric Defect Detection Techniques” Advances in Computational
Research, ISSN: 0975–3273, Vol.1, Issue 2, pp.18-29. 2009.
[5] S. Priya, T. Ashok Kumar and Paul Varghese, “A Novel Approach to
Fabric Defect Detection Using Digital Image Processing”, Proceedings
of International Conference on Signal Processing, Communication,
Computing and Networking Technologies (ICSCCN 2011),2011.

Authorized licensed use limited to: University of Exeter. Downloaded on June 15,2020 at 14:56:49 UTC from IEEE Xplore. Restrictions apply.
[6] X. F. Zhang and R. R. Bresee, “Fabric defect detection and classification [9] T. Ashok kumar, S.O. Priya and M.G. Mini, “Optic disc localization in
using image analysis”, Textile Res. J., Vol. 65, No.1, pp.1-9, 1995. ocular fundus images,” Proc. of iCVCi International Conference, India
[7] E. J. Cezar, “Applying Fourier and associated transforms to pattern 2011.
characterization in textiles,” Textile Res. J., Vol. 60, pp. 212-220, 1990. [10] T.J. Kang,“Automatic Structure Analysis and Objective Evaluation of
[8] C. Chan and G. K. H. Pang, “Fabric defect detection by Fourier Woven Fabric Using Image Analysis”, Textile Res. J. Vol.71,No.3,
analysis”, IEEE Trans. on Ind. Appl, Vol.36, No.5, pp.1267-1276, Oct pp.261-270, 2001.
2000.

Authorized licensed use limited to: University of Exeter. Downloaded on June 15,2020 at 14:56:49 UTC from IEEE Xplore. Restrictions apply.