2014c219jmalicirat 6 PDF
2014c219jmalicirat 6 PDF
net/publication/316476930
CITATIONS READS
3 908
3 authors:
Sakli Faouzi
Institut Supérieur des Etudes Technologiques de Ksar Hellal, Tunisia
427 PUBLICATIONS 1,367 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Optimization of technical characteristics and construction factors of Tunisian handmade carpets View project
All content following this page was uploaded by Sakli Faouzi on 26 April 2017.
CLASSIFICATION OF FABRICS DEFECTS USING IMAGE ANALYSIS AND FUZZY C‐MEANS METHOD
JMALI Mohamed, ZITOUNI Baghdadi and SAKLI Faouzi
Textile Engineering Laboratory ‐ University of Monastir, ISET Ksar Hellal, Tunisia.
ABSTRACT: Typically, the fabric inspection is performed by a human controller that uses a display system and
relies on personal knowledge. The objective of our work is to develop a system for the detection and
classification of defects in a simple and efficient way using techniques of image processing. Therefore, we
propose to provide an inspection process that aims to detect and classify defects in warp and weft using a
computer program developed in Matlab that analyzes images of fabrics samples acquired using a flat scanner.
KEYWORDS: Fabric defects, image analysis, fuzzy C‐means, classification techniques
1. INTRODUCTION
In (Xie, 2008), Xianghua Xie has shown that the
The textile industry is very concerned with surface defects are loosely separated into two
quality. It is increasingly imperative to produce types. One is local textural irregularities which is
goods of the highest quality in the shortest time the main concern for most visual surface
possible (Chen et al, 1998). inspection applications. The other is global
deviation of color and/or texture, where local
Among the new methods used to reduce the cost
pattern or texture does not exhibit abnormalities.
of poor quality, we find digital image processing
of fabrics (Chen et al., 2000). To recognize a In this paper, we begin by presenting the
sequence of forms, it is often necessary to different fabric defects treated (missing thread in
automatically segment the sequence before weft or warp, presence of an oil stain and
proceeding to the examination of each form presence of a hole in the fabric). After that, we
separately (Dorrity et al., 1994). deduce a discriminant character for each defect
(existence of straight lines of the same intensity,
Fabric defects are responsible for the majority of
uniformity of the processed image intensities and
problems found in the garment industry. It is
the existence of voids). Finally, we use the fuzzy
therefore imperative to detect and identify
C‐means method to classify defects.
defects to prevent them from reproducing (Japer
et al., 1995). Currently, much of the fabric 2. THE FABRICS DEFECTS
inspection is performed manually. With the most
In the manufacture of fabrics, many problems can
qualified inspectors, only some defects are
affect the textile. Among these anomalies, we
detected. In this context, there is a need for
find missing warp or weft threads, oil stains and
making an automatic inspection of fabrics in the
holes.
textile industry that can detect and identify the
characteristics of the fabric in real time. In this To detect and identify numerically these defects,
case, the system will alert maintenance they should be treated case by case. To do this,
personnel once the machine needs an we tried to find the characteristics that permit us
intervention to prevent the production of defects to assign each type of defect to its appropriate
or when it is necessary to modify the process category. The work aims to extract measurable
parameters to improve the quality of products parameters that can give us an idea of the
(Goswami et al., 2000). texture of the processed image in each case of
A new method of classifying fabrics defects has defect.
been designed and implemented to meet the 2.1. Missing thread in weft or warp
needs of manufacturers. To identify the most
harmful and expensive defects in fabrics, a survey The "missing thread in weft or warp" default is
was conducted in the industrial environment. very common in the weaving industry. In that
This survey showed that the most frequently case, the character of the texture is not respected
detected defects are missing weft or warp in some parts of the fabric. Figure 1 shows some
threads, oil stains and holes (Jeong et al., 2001). examples of this defect.
1
© CIRAT‐6, 2014
Figure 4: Distribution of samples according to the
presence of oil stains.
2.3. Presence of a hole in the fabric
During its displacement, the roll of fabric may
undergo some frictions which can cause more or
less large holes. Figure 5 shows some examples of
this defect.
Figure 2: Distribution of samples according to the
presence of straight lines.
2.2. Presence of an oil stain
During the operation of weaving, we have to
lubricate mechanical parts of the machine.
Therefore, the fabric may undergo some oil stains
accidentally. Figure 3 shows some examples of Figure 5: Examples of the defect "hole in the fabric."
this defect.
To calculate the rate of presence of void areas in
a fabric, we compute the percentage of white
pixels in the binary image. Then, we divide the
image into several sections and we determine the
percentage of white pixels in each section.
Finally, we sort the sections by the rate of white
pixels.
Figure 3: Examples of the defect "oil stain".
By applying this procedure to the database of
The main idea to detect oil stains is the division of samples having defects, we obtained the
the processed image into sections and calculating densities shown in figure 6. We notice that all
the percentage of black pixels in each section. samples having a hole present a high value for
This percentage is then compared to the this parameter.
2
the iteration number 11 the objective function is
equal to 0.797099 while for the iteration number
12, the objective function is equal to 0.797094.
Figure 6: Distribution of samples according to the
presence of voids.
3. CLASSIFICATION BY FUZZY C‐MEANS METHOD
Fuzzy C‐Means (FCM) is a fuzzy clustering
algorithm. It introduces the concept of fuzzy set Figure 8: The progression of the classification.
in the class definition: each point in the data set 3.3. The cluster centers
for each group with a certain degree of
The knowledge of cluster centers allows us to
membership. In fuzzy clustering, every point has
determine the characteristics of each class in
a degree of belonging to clusters, as in fuzzy logic,
each default.
rather than belonging completely to just one
cluster. Thus, points on the edge of a cluster, may We refer in Table 1 the classes and the
be in the cluster to a lesser degree than points in parameters as follows:
the center of cluster. Class 1: default "missing wire"
3.1. Classification of samples
Class 2: default "hole"
Since all the data belong to three classes that
Class 3: default "oil stain"
correspond to the three defaults studied: missing
thread, oil stain and hole in the fabric, we PLC: percentage which indicates the presence of
presented the distribution of all samples into continuous lines.
three classes. Figure 7 shows the distribution of PZV: percentage that indicates the presence
fabrics in 3 classes whose centers are marked in of voids.
bold.
PZS: percentage that indicates the presence
1
of dark areas.
0.9
Table 1 : The cluster centres.
0.8
3
© CIRAT‐6, 2014
Table 1 : The degree of membership of the samples which is very high compared to the degrees of
REF DEGREES OF MEMBERSHIP belonging to other classes.
TISSU CLASSE1 CLASSE2 CLASSE3 4. CONCLUSION
T1 0,9119 0,0380 0,0501
Quality is a very important factor in textile
T2 0,7717 0,0879 0,1403 industry. In our work, we tried to implement a
T3 0,9007 0,0513 0,0479 system based on image processing and Fuzzy C‐
T4 0,5888 0,1708 0,2405 means to detect and classify three fabric defects
T5 0,9846 0,0076 0,0078 (missing threads, oil stains and holes).
T6 0,5935 0,2798 0,1267
T7 0,9968 0,0016 0,0015 Our system works according to four steps. We
T8 0,8921 0,0556 0,0524 begin by capturing the image and then we
eliminate parasite information and increase the
T9 0,7308 0,1224 0,1468
sharpness of the image by applying appropriate
T10 0,9347 0,0336 0,0318
preprocessing operations. After that, we
T11 0,7010 0,1530 0,1460
determine three parameters witch characterize
T12 0,7301 0,1145 0,1553
the mentioned defects (the rate of straight lines,
T13 0,8734 0,0753 0,0513
the rate of dark areas and the rate of voids).
T14 0,9480 0,0282 0,0238 Finally we use a Fuzzy C‐means method to
T15 0,9370 0,0360 0,0270 recognize and classify the category of defect
T16 0,0797 0,8158 0,1045 present on the fabric.
T17 0,0518 0,8566 0,0916
T18 0,1040 0,7482 0,1479 Our system can be improved to recognize other
T19 0,0807 0,8322 0,0871 types of defects in real time and treat fabrics
T20 0,0471 0,8054 0,1475 which contain more than one defect.
T21 0,0456 0,8542 0,1002 REFERENCES
T22 0,0971 0,7909 0,1120 Chen. P. W, Liang. T. C., , and Yau. H. F (1998):
T23 0,0462 0,8305 0,1233 “Classifying Textile Faults with a Back‐Propagation
T24 0,1070 0,6834 0,2095 Neural Network Using Power Spectra”, Textile
T25 0,1427 0,7590 0,0983 Research Journal, 68 (2), pp. 121‐126.
T26 0,0622 0,8749 0,0629 Dorrity. J, Vachtsevanos. G and Jasper. W (1994):
T27 0,0532 0,8844 0,0623 “Real‐time fabric defect detection and control in
T28 0,0798 0,6827 0,2375 weaving process” ; Nat Textil Center Annu Rep 1994,
T29 0,0819 0,7539 0,1642 20, pp. 193–202.
T30 0,0879 0,8094 0,1027 Goswami. B and Datta. A. k. (2000): “Detecting
T31 0,0643 0,1248 0,8109 Defects in Fabric with Laser‐Based Morphological
T32 0,1910 0,1713 0,6377 Image Processing”, Textile Research Journal, 70(9),
T33 0,0106 0,0133 0,9761 pp. 758‐762.
T34 0,0240 0,0344 0,9417 Jasper. W et Potlapalli. H (1995): “Image Analysis of
T35 0,0273 0,0565 0,9162 Mispicks in Woven Fabric”, Textile Research Journal,
T36 0,0475 0,0455 0,9070 65(11), pp. 683‐692.
T37 0,0386 0,0775 0,8839 Jeong. S, Choi. H, Kim. S, Jaung. J.Y and Kim. S. H
T38 0,0775 0,0986 0,8239 (2001): “Detecting Fabric Defects with Computer
T39 0,0457 0,1346 0,8198 Vision and Fuzzy Rule Generation. Part I: Defect
T40 0,0105 0,0169 0,9726 Classification by Image Processing”, Textile Research
T41 0,1737 0,1131 0,7131 Journal, 71(6), pp. 518‐526.
T42 0,1418 0,1595 0,6987 Xie. X (2008): “A Review of Recent Advances in
T43 0,2367 0,5105 0,2528 Surface Defect Detection using texture analysis
T44 0,1232 0,1873 0,6896 Techniques”, Electronic Letters on Computer Vision
T45 0,1125 0,2887 0,5988 and Image Analysis, 7(3), pp. 1‐22.
4