https://doi.org/10.31449/inf.v46i1.3015 Informatica 46 (2022) 19–25 19 
 
Automatic Fabric Inspection using GLCM-based Jensen-Shannon 
Divergence 
V. Asha
 
E-mail: v_asha@live.com; asha.gurudath@gmail.com 
Department of Master of Computer Applications, New Horizon College of Engineering, Bangalore, India 
ORCID: 0000-0003-4803-099X 
Keywords: cluster, co-occurrence matrix, defect, periodicity, Jensen-Shannon, divergence  
Received: November 25, 2019 
Jensen-Shannon divergence is one of the powerful information-theoretic measures that can capture 
mutual information between two probability distributions. In this paper, a machine vision algorithm is 
proposed for automatic inspection on dot patterned fabric using Jensen-Shannon divergence based on 
gray level co-occurrence matrix (GLCM). Input defective images are split into several periodic blocks 
based on their periodicities extracted using superposition of distance matching functions and the gray 
levels are quantized from 0-255 to 0-63 to keep the GLCM compact and to reduce the computation time. 
Symmetric Jensen-Shannon divergence metrics are calculated from the GLCMs of each periodic block 
with respect to itself and all other periodic blocks to get a dissimilarity matrix that satisfies true metrics 
conditions. This dissimilarity matrix is subjected to hierarchical clustering to automatically identify 
defective and defect-free blocks. Results from experiments on real fabric images with defects such as 
broken end, hole, thin bar, thick bar, netting multiple and knot show the effectiveness of the proposed 
method for fabric inspection. 
Povzetek: Razvita je metoda, ki na osnovi Jensen-Shannovove divergence omogoča iskanje pokvarjenih 
vzorcev na ponavljajoči se shemi. 
1 Introduction 
Periodically patterned texture images are often found in 
day-to-day life in various applications such as ceramic 
tiles, wallpapers, and textile fabrics. Inspection of these 
products plays a major role with regard to quality control 
of the products in industries. Conventional human-vision 
based inspection are being followed in many of the 
industries. Lack of repeatability and reproducibility of 
inspection results due to fatigue and subjective nature of 
humans and imperfect inspection due to complicated 
design in the texture patterns are common in the 
conventional human-vision based inspection systems. 
Automated machine-vision based inspection can promote 
the production rate of the products by decreasing the 
inspection time compared to the traditional human-vision 
based inspection especially in industries such as textile 
industries. Periodic textures which we come across in our 
life include several natural patterns and made-made / 
artificial patterns. Plenty of patterns are used in textile 
fabric design. These patterns, in turn, include either 
random structures or periodic structures, making the 
fabric appealing to our eyes. Inspection on fabric images 
having periodic texture patterns is more complicated than 
that on plain and twill fabric images due to complexity in 
the design, existence of numerous categories of patterns, 
and similarity between the defect and background [1]. 
For the inspection on periodically patterned fabrics, there 
are methods in literature that depend on training stage 
with numerous defect-free samples for obtaining 
decision-boundaries or thresholds prior to detection of 
defects [1]-[11]. However, unsupervised method of 
identifying defects in a patterned texture image is quite 
challenging [12]-[17]. In this paper, a method of 
inspection on periodically patterned fabrics is proposed 
without any training stage with the help of texture-
periodicity and Jensen-Shannon divergence metric based 
on gray level co-occurrence matrix. The main 
contributions of this research can be summarized as 
follows: 
• There is no training stage with defect-free samples 
for decision boundaries or thresholds unlike other 
methods. 
• Due to the absence of training stage with defect-
free samples, the proposed method does not need 
huge memory space for storage of defect-free 
samples. 
• Identification of defective and defect-free periodic 
units of the fabrics is automatically carried out 
based on cluster analysis without human 
intervention. 
The program for the proposed algorithm is written in 
Matlab-7.0. The organization of this paper is as follows: 
Section-2 presents a brief review on Jensen-Shannon 
divergence, gray level co-occurrence matrix and the 
proposed algorithm for fabric inspection along with 
illustration and results from experiments on various real 
fabric images with defects. Section-3 has the 
conclusions. 

20 Informatica 46 (2022) 19–25 V. Asha 
 
2 Proposed algorithm for fabric 
inspection 
Jensen-Shannon divergence is one of the effective 
measures for capturing mutual information between two 
probability distributions [12]. In this paper, it is intended 
to employ Jensen-Shannon divergence metric to 
discriminate between defective and defect-free periodic 
blocks as feature extraction (e.g. [18],[19]). Following 
the fact that human vision perception makes use of 
second-order statistics for texture discrimination [20], 
histogram constructed from second-order statistics is 
used to discern between defective and defect-free 
periodic blocks. Since the time of proposal of 14 features 
derived from histogram of second order statistics or gray-
level co-occurrence matrices (GLCM) by Haralick et al. 
[21], several authors have employed GLCM features for 
various texture analysis applications (e.g.[22]-[26]). 
Instead of extracting GLCM features, information from 
every pixel-pair of a periodic block with that of other 
periodic block is used to compute a distance metric using 
Jensen-Shannon divergence. The uniqueness of this 
Jensen-Shannon divergence metric based on GLCM is 
that the distance measure is considered for every pair of 
gray levels between two periodic blocks unlike the 
conventional GLCM features that are computed for the 
entire GLCM. 
2.1 Brief review on Jensen-Shannon 
divergence  
Jensen-Shannon divergence is a symmetrized, smoothed 
version of Kullback-Leibler divergence which is the most 
important divergence measure of information theory 
[18]. Given two classes p(  1) and p(  2) with a common 
feature vector x, the Kullback-Leibler divergence or the 
relative entropy in terms of ratio of the probability 
distributions of these classes conveys useful information 
concerning the capability of discriminating the two 
classes. This class-separability measure over class  1 is 
given as [27]  
(𝜔 1
,𝜔 2
)=∫ 𝑝 (𝜔 1
)𝑙𝑜𝑔 (
𝑝 (𝜔 1
)
𝑝 (𝜔 2
)
)𝑑𝑥
∞
−∞
  (1) 
Similarly, the class-separability measure over class  2 is 
given as 
(𝜔 2
,𝜔 1
)=∫ 𝑝 (𝜔 2
)𝑙𝑜𝑔 (
𝑝 (𝜔 2
)
𝑝 (𝜔 1
)
)𝑑𝑥
∞
−∞
  (2) 
Based on the probability distributions p( 1) and p(  2) 
and their average, a symmetric divergence called Jensen-
Shannon divergence  is given as 
=
∫ {𝑝 (𝜔 1
)log(
0.5×𝑝 (𝜔 1
)
𝑝 (𝜔 1
)+𝑝 (𝜔 2
)
)+𝑝 (𝜔 2
)log(
0.5×𝑝 (𝜔 2
)
𝑝 (𝜔 1
)+𝑝 (𝜔 2
)
)𝑑𝑥 }
∞
−∞
      (3) 
Since Kullback-Leibler divergence  (𝜔 1
,𝜔 2
) or  
(𝜔 1
,𝜔 2
) can be interpreted as the inefficiency of 
assuming that the true distribution is p(𝜔 2
) or p(𝜔 1
) 
when it is really p(𝜔 1
) or p(𝜔 2
), Jensen-Shannon 
divergence can be considered as a minimum inefficiency 
distance. 
2.2 Brief review on Gray Level Co-
occurrence Matrix 
Gray level co-occurrence matrices (GLCM) have been 
employed in extracting texture features for various 
applications for a long time. It is a measure of gray-level 
dependencies in a local neighborhood for a given pixel 
displacement and orientation. In other words, a GLCM is 
a matrix showing the number of times a pixel with gray 
level I occurs at position vector from a pixel with gray 
level j. Mathematically, the GLCM for an imageI(x, y) of 
size (M, N) at a given offset ( x, y) is given as 
𝐶 x, y
=∑ ∑ {
1,𝑖𝑓 𝐼 (𝑥 ,𝑦 )=𝑖 𝑎𝑛𝑑 
𝐼 (𝑥 +Δ𝑥 ,𝑦 +Δ𝑦 )=𝑗 ;
0,𝑜𝑡 ℎ𝑒𝑟𝑤𝑖𝑠𝑒 .
𝐿 −1
𝑗 =0
𝐿 −1
𝑖 =0
 (4) 
where, L is the total number of gray values in the image. 
For an-bit image, the value of L is 2n with dynamic 
range of gray values being [0, L-1]. The offset ( x, y) 
characterizes the pixel displacement and the orientation 
for the co-occurrence matrix. 
2.3 Description of the algorithm 
There are three major assumptions in the proposed 
algorithm as follows: 
• Test image contains at least two periodic units in 
horizontal direction and two in vertical direction whose 
row and column dimensions are known a priori. 
• Number of defective periodic units is always less 
than the number of defect-free periodic units. 
• Test images are from imaging system oriented 
perpendicular to the surface of the product to be 
inspected. This assumption is due to the fact that in an 
inspection system in industries, the imaging system is 
always oriented perpendicular to the plane of the surface 
of the products such as tiles, wallpapers, and textile 
fabrics. 
An image under inspection may have fractional 
periods also. Hence, the concept of analyzing the 
periodic blocks extracted from all four corners of the test 
image is used for identifying the defects followed by the 
concept of defect fusion proposed in [28]. Four cropped 
images are obtained from the defective test image by 
cropping the input image from all four corners (top-left, 
bottom-left, top-right and bottom-right) to get complete 
number of periodic blocks. If g is an image of size M × N 
with row periodicity P r (i.e., number of columns in a 
periodic unit) and column periodicity P c (i.e., number of 
rows in a periodic unit), size of the cropped images is 
M crop × N crop, where M crop and N crop are measured from 
top-left, bottom-left, top-right and bottom-right corners 
and are given by the following equations: 
𝑀 𝑐𝑟𝑜𝑝 =𝑓𝑙𝑜𝑜𝑟 (
𝑀 𝑃 𝑐 ) 𝑃 𝑐   (5) 
𝑁 𝑐𝑟𝑜𝑝 =𝑓𝑙𝑜𝑜𝑟 (
𝑁 𝑃 𝑟 ) 𝑃 𝑟   (6) 
Each cropped image is split into several periodic 
blocks of size Pc x Pr. The periodicities are estimated 
using Superposition of distance matching function 
proposed in [29]. The decision that one has to make now 

Automatic Fabric Inspection using GLCM-based ... Informatica 46 (2022) 19–25 21 
 
is how many levels are needed to construct a GLCM for 
effective texture analysis. The number of gray levels is 
an important factor in the computation of GLCM. The 
more levels included in the computation, the more 
accurate the extracted textural information, with, of 
course, a subsequent increase in computation costs and 
too less gray levels will result in loss of information due 
to quantization. In general, the effect of false-contouring 
starts predominating in an image if the gray levels are 
quantized below the dynamic range of gray values 0-63 
[30]. In the proposed method, each 8-bit test image 
having a dynamic range of gray values 0-255 is linearly 
quantized to 6-bit image having a dynamic range 0-63 
and rotation-invariant GLCMs (sum of GLCMs over 8 
directions θ € {0, /4, /2, 3 /4, , 5 /4, 3 /2, 7 /8}) 
calculated for a unit pixel displacement is utilized. If p(i, 
j) and q(i, j) represent the summed up GLCMs of two 
periodic blocks A and B respectively whose dynamic 
range of gray values is between 0 and L-1 for a unit pixel 
displacement, the Jensen-Shannon divergence can be 
computed as  
Λ
𝐴 ,𝐵 =Λ
𝐵 ,𝐴 =
 ∑ ∑ {𝑝 (𝑖,𝑗)log(
0.5×𝑝 (𝑖 ,𝑗 )
𝑝 (𝑖 ,𝑗 )+𝑞 (𝑖 ,𝑗 )
)+𝑞 (𝑖,𝑗)log(
0.5×𝑞 (𝑖 ,𝑗 )
𝑝 (𝑖 ,𝑗 )+𝑞 (𝑖 ,𝑗 )
)}
𝐿 −1
𝑗 =0
𝐿 −1
𝑖=0
      (7) 
The square root of this divergence is a true metric 
obeying the conditions of a true metric in 2D Euclidean 
space [18], viz.,
√

𝑖 ,𝑗  0 for all i and j (non-negativity), 
√

𝑖 ,𝑗 = 0 for all i = j (self-distance) and 
√

𝑖 ,𝑗 = 
√

𝑗 ,𝑖 for 
all i and j (symmetry), and 
√

𝑖 ,𝑗 ≤ √ 
𝑖 ,𝑘 + 
√

𝑘 ,𝑗 for i ≠ j 
≠ k (triangular inequality). Following Eq. (7) based on 
Jensen-Shannon divergence for each periodic block with 
respect to itself and all other periodic blocks, a 
dissimilarity matrix D containing true distance metrics 
can be obtained as below: 
𝐷 =
(
 
 
 
√Λ
1,1
√Λ
1,2
... √Λ
1,𝑛 −1
√Λ
1,𝑛 √Λ
2,1
√Λ
2,2
⋯ √Λ
2,𝑛 −1
√Λ
2,𝑛 ⋮ ⋮ ⋮ ⋮ ⋮
√Λ
𝑛 −1,1
√Λ
𝑛 −1,2
… √Λ
𝑛 −1,𝑛 −1
√Λ
𝑛 −1,𝑛 √Λ
𝑛 ,1
√Λ
𝑛 ,2
⋯ √Λ
𝑛 ,𝑛 −1
√Λ
𝑛 ,𝑛 )
 
 
 
 (8) 
It should be noted that if a cropped image has n 
number of periodic blocks, then the size of the 
dissimilarity matrix is n x n. Because the divergence 
measure of a periodic block with itself is zero and the 
divergence measure between ith periodic block and jth 
periodic block is same as the divergence measure 
between jth periodic block and ith periodic block, the 
dissimilarity matrix becomes a diagonally symmetric 
matrix with diagonal elements being zero as below: 
𝐷 =
(
 
 
 
0
√Λ
2,1
0
⋮ ⋮ ⋮
√Λ
𝑛 −1,1
√Λ
𝑛 −1,2
… 0
√Λ
𝑛 ,1
√Λ
𝑛 ,2
⋯ √Λ
𝑛 ,𝑛 −1
0
)
 
 
 
 (9) 
It may be noted that because the matrix is similar 
about the diagonal, the upper diagonal elements are not 
filled for the sake of simplicity. This dissimilarity matrix 
is directly given as input to the Ward's cluster algorithm 
[28],[31] to automatically get defective and defect-free 
periodic blocks from each cropped image. Detection of 
defective periodic blocks from each cropped image does 
not give an overview of the total defects in the input 
defective image. Hence, in order to get the overview of 
the total defects in the input image, defect-fusion 
proposed in [28] is used which involves fusion of the 
boundaries of the defective periodic blocks identified 
from each cropped image, morphological filling and 
Canny edge extraction. 
2.4 Illustration of the algorithm 
In order to illustrate the proposed algorithm for fabric 
inspection, let us a defective box-patterned fabric image 
as shown in Fig. 1 (a). Following Eqs. (5) and (6), four 
cropped images containing complete number of periodic 
blocks are obtained as shown in Fig. 1 (b) - (e), from the 
test image with the help of periodicities known a priori.  
Each cropped image is split into several blocks of 
size same as the size of the periodic unit and the gray 
values are quantized from 0-255 to 0-63. Jensen-Shannon 
divergence metrics are calculated for each periodic block 
with respect to itself and all other periodic blocks and a 
dissimilarity matrix is obtained from the GLCMs 
calculated over 8 directions for unit pixel displacement. 
The dissimilarity matrices thus obtained are shown in 
Fig. 2 in gray-scale form by scaling the matrix elements 
linearly in the range 0-255. Dark pixels indicate 
closeness among the periodic blocks and bright pixels 
indicate high dissimilarity. It may be noted from Fig. 2 
that the diagonal elements in the dissimilarity matrix 
clearly indicate that the periodic blocks are of zero 
dissimilarity with themselves and that the dissimilarity 
matrix is symmetric. 
 
Figure 1: (a) Input defective image; (b) Cropped image 
obtained from top-left corner of the input image; (c) 
Cropped image obtained from bottom-left corner of the 
input image; (d) Cropped image obtained from top-right 
corner of the input image; (e) Cropped image obtained 
from bottom-right corner of the input image. 
 
Figure 2: Dissimilarity matrix derived from the Jensen-
Shannon divergence metrics for the cropped image 
obtained from (a) top-left (b) bottom-left (c) top-right 
and (d) bottom-right corners of the test image shown in 
gray-scale. 

22 Informatica 46 (2022) 19–25 V. Asha 
 
The dendrograms obtained from the dissimilarity 
matrices through Ward's hierarchical clustering are 
shown in Fig. 3 along with the defective periodic blocks 
identified by the clustering for all cropped images. 
Boundaries of the defective periodic blocks thus 
identified from each cropped image are highlighted using 
white pixels and shown in Fig. 4. Boundaries of the 
defective periodic blocks identified from each cropped 
image are shown superimposed on the original image in 
Fig. 5 (a) and separately on a plain background in Fig. 5 
(b). Fig. 5 (c) shows the result of morphological filling 
and Fig. 5 (d) shows the edges of the total defects 
superimposed on the original defective image after 
Canny edge detection. Thus, an overview of the total 
defects on the original image itself can be obtained 
following the concept of fusion, morphological filling 
and edge detection. It may be noted that though the 
number of periodic blocks taken from a defective input 
image is same for all of its cropped images, the number 
of defective periodic blocks identified does not need to 
be same for all cropped images. This is because the 
contribution of defect in each periodic block may differ 
for different cropped images.  Nevertheless, fusion of 
defects from all 4 cropped images helps in getting an 
overview of total defects in the input image. This is 
clearly evident from the example image used for 
illustration of the proposed algorithm. 
2.5 Experiments on real fabric images with 
different types of defects 
Periodically patterned fabric images with defects such as 
broken end, thin bar, thick bar, netting multiple and knot 
are tested to study the performance of the proposed 
algorithm of fabric inspection. Fig. 6 shows the defective 
fabric images and Fig. 7 shows the final result of defect 
identification after fusion, morphological filling and edge 
detection. 
2.6 Performance evaluation and 
comparison with other methods 
Precision, recall and accuracy are the widely used 
performance parameters in several retrieval applications. 
In order to access the performance of the proposed 
method, these performance parameters are evaluated in 
terms of true positive (TP), true negative (TN), false 
positive (FP), and false negative (FN), [31], [33], [34], 
[35]. True positive is defined as the number of defective 
periodic blocks identified as defective. True negative is 
defined as the number of defect-free periodic blocks 
 
Figure 3: Dendrogram obtained from cluster analysis of 
the matrix containing Jensen-Shannon divergence 
metrics obtained from the test image by cropping it from 
(a) top-left (b) bottom-left (c) top-right and (d) bottom-
right corners. Defective blocks identified from these 
cropped images are (8, 16, 27, 9, 17, 28, 24, 35, 53, 15, 
34, 52, 22, and 23), (8, 16, 27, 24, 35, 53, 9, 17, 28, 15, 
34, 52, 22, and 23), (28, 46, 17, 9, 8, 27, 45, 15, 16, 1, 
20, 2, 10, and 21) and (28, 46, 17, 9, 8, 27, 45, 15, and 
16).  
 
Figure 4: Defective periodic blocks identified from the 
cluster analysis of dissimilarity matrix derived from the 
Jensen-Shannon divergence metrics of the cropped 
images obtained from (a) top-left (b) bottom-left (c) top-
right and (d) bottom-right corners of the test image with 
their boundaries highlighted using white pixels.  
 
Figure 5: Defective periodic blocks identified from the 
cluster analysis of dissimilarity matrix derived from the 
Jensen-Shannon divergence metrics of the cropped 
images obtained from (a) top-left (b) bottom-left (c) top-
right and (d) bottom-right corners of the test image with 
their boundaries highlighted using white pixels. 
Illustration of defect fusion: (a) Boundaries of the 
defective blocks identified from each cropped image 
shown super-imposed on the original image; (b) 
Boundaries of the defective blocks shown separately on 
plain background; (c) Result of morphological filling; (d) 
Edges identified through Canny edge operator shown 
superimposed on original defective image using white 
pixels.  
 
Figure 6: Sample fabric images with defect - (a) broken 
end, (b) hole, (c) thin bar, (d) thick bar, (e) netting 
multiple, and (f) knot. 

Automatic Fabric Inspection using GLCM-based ... Informatica 46 (2022) 19–25 23 
 
identified as defect-free. False positive is defined as the 
number of defect-free periodic blocks identified as 
defective. False negative is defined as the number of 
defective periodic blocks identified as defect-free. 
Precision is defined as the number of periodic blocks 
correctly labelled as belonging to the positive class 
divided by the total number of periodic blocks labelled as 
belonging to the positive class and is calculated as 
TP/(TP+FP). Recall is defined as the number of true 
positives divided by the sum of true positives and false 
negatives that are periodic blocks not labelled as 
belonging to the positive class but should have been and 
is calculated as TP/(TP+FN). Accuracy is the measure of 
success rate that takes into account the detection rates of 
defective and defect-free periodic blocks and is 
calculated as (TP+TN)/(TP+TN+FP+FN). These 
performance parameters are averaged for all cropped 
images and given in Table 1. The average precision, 
recall and accuracy for all the defective images (based on 
a total of 1764 number of periodic blocks) are 100%, 
83.6%, and 97.2% respectively. The accuracy of a few 
other methods available in literature for defect detection 
on patterned textures varies from 88% to 99% [36]. 
Relatively less recall rates in the proposed method 
indicate that there are few false negatives identified by 
the proposed method. However, as the proposed method 
yields high precision and accuracy, the proposed method 
can contribute to automatic inspection in industries such 
as fabric industries. In general, the computational time 
complexity of a full GLCM is O(M
2
N
2
). As the proposed 
method is based on reduced GLCM from 0-255 to 0-63, 
the time complexity is reduced by one-third. 
3 Conclusion 
In this paper, texture-periodicity and Jensen-Shannon 
divergence metrics have effectively been used for the 
development of the automated inspection on periodically 
patterned fabrics. Absence of training stage with defect-
free test samples for obtaining decision-boundaries or 
thresholds, unsupervised method of identifying defects 
using cluster analysis, and high success rates are the 
novelties of the proposed method. Thus, the proposed 
method can contribute to the development of 
computerized inspection and quality control in industries 
such as fabric industries.  
Acknowledgement 
The author would like to thank Dr. Henry Y. T. Ngan, 
Research Assistant Professor, Department of 
Mathematics, Hong Kong Baptist University, Kowloon, 
for providing the database of patterned fabrics. 
References 
[1] H.Y.T Ngan, Pang and G.H.K. Regularity Analysis 
for Patterned Texture Inspection. IEEE Trans.  
Autom. Sci. Eng., Vol. 6 (1) (2009), pp. 131-144. 
(DOI: 10.1109/TASE.2008.917140) 
[2] H.Y.T Ngan, Pang, G.H.K. and N.H.C Yung. 
Performance Evaluation for Motif-Based Patterned 
Texture Defect Detection. IEEE Trans. Autom. Sci. 
Eng. Vol. 7 (1) (2010) pp. 58-72.  
(DOI:10.1109/TASE.2008.2005418) 
[3] H.Y.T Ngan, Pang and G.H.K.  Novel method for 
patterned fabric inspection using Bollinger bands. 
Opt. Eng. Vol. 45 (8) (2006), pp. 087202-1-15.  
(DOI:10.1117/1.2345189) 
[4] F Tajeripour, E Kabir and A Sheikhi.  Fabric Defect 
Detection Using Modified Local Binary Patterns. 
In: Proc. of the Int. Conf. on Comput. Intel. And 
Multimed. Appl., (2007) pp. 261-267.  
(https://doi.org/10.1155/2008/783898) 
[5] H.Y.T Ngan, Pang, G.H.K and N.H.C Yung. Motif-
based defect detection for patterned fabric. Pattern 
 
Figure 7: (a) Boundaries of the defective blocks 
identified from each cropped image shown super-
imposed on the original image; (b) Boundaries of the 
defective blocks shown separately on plain background; 
(c) Result of morphological filling; (d) Edges identified 
through Canny edge operator shown superimposed on 
original defective image using white pixels. Final result 
of defect identification after defect fusion, morphological 
filling and edge detection for (a) defective fabric image 
with broken end; (b) defective fabric image with hole; (c) 
defective fabric image with thin bar; (d) defective fabric 
image with thick bar; (e) defective fabric image with 
netting multiple; (f) defective fabric image with knot. 
Table 1: Summary of performance parameters of the 
proposed algorithm. 
Image defect Precision 
(%) 
Recall 
(%) 
Accuracy 
(%) 
Hole  
(in illustration) 
100 82.1 95.6 
Broken end 100 80.0 96.8 
Hole 100 85.7 98.4 
Thin bar 100 87.5 99.2 
Thick bar 100 91.7 97.6 
Netting multiple 100 77.5 97.2 
Knot 100 80.4 95.6 
Average 100 83.6 97.2 
 

24 Informatica 46 (2022) 19–25 V. Asha 
 
Recognit. Vol. 41, (2008), pp.1878-1894.  
(https://doi.org/10.1016/j.patcog.2007.11.014) 
[6] H.Y.T Ngan, Pang, G.H.K. and N.H.C Yung. 
Ellipsoidal decision regions for motif-based 
patterned fabric defect detection. Pattern Recognit. 
Vol. 43, (2010) pp.2132-2144.  
(DOI:10.1016/j.patcog.2009.12.001) 
[7] M.K Ng, H.Y.T Ngan, X.Yuan  and W. Zhang.  
Patterned Fabric Inspection and Visualization by 
the Method of Image Decomposition. IEEE Trans. 
Autom. Sci. Eng. Vol. 11 (3) (2014), pp. 943-947. 
(DOI: 10.1109/TASE.2014.2314240) 
[8] J Liang, C Chen, L. Jiuzhen, L and H Zhenjie. 
Fabric defect inspection based on lattice 
segmentation and Gabor filtering. Neurocomput. 
Vol. 238 (5), (2017), pp. 84-102.   
(https://doi.org/10.1016/j.neucom.2017.01.039) 
[9] S Mei, Y. Wang, and G. Wen., Automatic Fabric 
Defect Detection with a Multi-Scale Convolutional 
Denoising Auto-encoder Network Model. Sensors. 
Vol. 18 (4), (2018), 1064-1074.  
(DOI:10.3390/s18041064) 
[10] C. Li, G. Gao, Z. Liu, D. Huang and J  Xi. Defect  
Detection for Patterned Fabric Images Based on 
GHOG and Low-Rank Decomposition. IEEE 
Access, Vol 7, (2019), pp. 83962-83973.   
DOI: 10.1109/ACCESS.2019.2925196 
[11] R.A.L Morales, R.E.S Yanez and R.B. Serrato. 
Defect detection on patterned fabrics using texture 
periodicity and the coordinated clusters 
representation, Textile Research J., Vol 87 (15), 
(2016), pp.  1869-1882.  
https://doi.org/10.1177/0040517516660885 
[12] V. Asha, N.U. Bhajantri and P. Nagabhushan.  
Automatic Detection of Defects on Periodically 
Patterned Textures, Journal of Intelligent Systems, 
vol. 20 (3), (2011), pp. 279-303.   
(https://www.degruyter.com/document/doi/10.1515/
jisys.2011.015/html) 
[13] V. Asha, N.U. Bhajantri and P. Nagabhushan. 
GLCM-based chi-square histogram distance for 
automatic detection of defects on patterned textures. 
International Journal of Computational Vision and 
Robotics (IJCVR), vol. 2 (4), (2011), pp. 302-313.  
(DOI: 10.1504/IJCVR.2011.045267) 
[14] V. Asha, N.U. Bhajantri and P. Nagabhushan. 
Automatic Detection of Texture-defects using 
Texture-periodicity and Jensen-Shannon 
Divergence. Journal of Information Processing 
Systems, vol. 8 (2), (2012) pp. 359-374, 2012.   
(https://doi.org/10.3745/JIPS.2012.8.2.359) 
[15] V. Asha, N.U. Bhajantri and P. Nagabhushan. 
Similarity measures for automatic defect detection 
on patterned textures. International Journal of 
Information and Communication Technology, vol. 
4 (2/3/4), (2012), pp. 118-131.   
(https://www.inderscienceonline.com/doi/abs/10.15
04/IJICT.2012.048758) 
[16] V. Asha. Singular Value Decomposition of Images 
of Structural Textures and their Reconstruction 
using Periodicity. International Journal of 
Tomography and Simulation (IJTS), vol 31, Issue 4, 
(2018), pp 86-97.  
(http://www.ceser.in/ceserp/index.php/ijts/article/vi
ew/5696) 
[17] V. Asha. Texture Defect Detection using Human 
Vision Perception based Contrast. International 
Journal of Tomography and Simulation (IJTS), vol 
32, Issue 3, (2019), pp 86-97.  
(http://www.ceser.in/ceserp/index.php/ijts/article/vi
ew/6017) 
[18] H. Jiang, Defect Features Recognition in 3D 
Industrial CT Images, Informatica – An Int. J. of 
Comput. and Inform., Vo. 42, No. 3 (2018). 
(https://doi.org/10.31449/inf.v42i3.2454) 
[19] D.M Endres and J.E Schindelin. A New Metric for 
Probability Distributions. IEEE Trans. on Info. 
Theory, Vol 49 (7), (2003), pp.1858-1860.  
(DOI: 10.1109/TIT.2003.813506) 
[20] T Caelli, B. Julesz and E. Gilbert. On Perceptual 
Analyzers Underlying Visual Texture 
Discrimination, Part II. Biol. Cybern. Vol 29(4), 
(1978), pp. 201-214.  
(DOI: 10.1007/BF00337138) 
[21] R.M. Haralick, K. Shanmugam and I. Dinstein. 
Textural features for image classification. IEEE 
Trans. On Syst., Man and Cybern. Vol 3(6), (1973), 
pp. 610-621.  
(DOI:10.1109/TSMC.1973.4309314) 
[22] R.M Haralick. Statistical and structural approaches 
to texture. In: Proc. of the IEEE, Vol 67(5), (1979) 
pp. 786 – 804.  
(DOI: 10.1109/PROC.1979.11328) 
[23] A.L Amet, A. Ertuzun and A. Ercil.  Texture defect 
detection using subband domain co-occurrence 
matrices. Image and Vis. Comp. Vol 18, (2000), pp. 
543-553. 
(DOI: 10.1109/IAI.1998.666886) 
[24] Fernandos, J.C.A., Neves, J.A.B.C, Couto and 
C.A.C. Defect Detection and Localization in 
Textiles using Co-occurrence Matrices and 
Morphological Operators. In:Proc. of the 6th Int. 

Automatic Fabric Inspection using GLCM-based ... Informatica 46 (2022) 19–25 25 
 
Conf. onMechatron. Mach. Vis. In Pract. 
(M2VIP'99), Ankara, Turkey. (1999) 
[25] C.F.J Kuo and T.L Su.  Gray Relational Analysis 
for Recognizing Fabric Defects. Textile Res. J. Vol 
73(5), (2003), pp. 461-465   
(https://doi.org/10.1177/004051750307300515) 
[26] L.H Soh and C Tsatsoulis. Texture Analysis of SAR 
SeaIce Imagery Using Gray Level Co-occurrence 
Matrices. IEEE Trans. On Geosci. Remote Sens. 
Vol 37(2), (1999), pp. 780-795.  
(DOI: 10.1109/36.752194) 
[27] S. Theodoridis and K.  Koutroumbas, Pattern 
Recognition, Fourth Edition, Academic Press, CA 
(2009) 
[28] V. Asha, N.U. Bhajantri and P. Nagabhushan. 
Automatic Detection of Texture Defectsusing 
Texture-Periodicity and Gabor Wavelets. In: 
Venugopal, K.R., and Patnaik, L.M. (Eds.): ICIP 
2011, Communication and Computer Information 
Series (CCIS) 157, Springer-Verlag, Berlin 
Heidelberg, (2011), pp. 548-553. 
[29] V. Asha, P. Nagabhushan and N.U. Bhajantri. 
Automatic extraction of texture-periodicity using 
superposition of distance matching functions and 
their forward differences. Pattern Recog. Lett. 33 
(5), (2012), pp. 629-640.  
(https://doi.org/10.1016/j.patrec.2011.11.027) 
[30] R.C. Gonzalez, R.E. Woods.  Digital Image 
Processing. Pearson Prentice Hall, New Delhi. 
(2008) 
[31] J. Pisanki and T. Pisanki, The use of collaboration 
distance in scheduling conference talks, Informatica 
– An Int. J. of Comput. and Inform., Vo. 43, No. 4 
(2019). (https://doi.org/10.31449/inf.v43i4.2832) 
[32] T. Fawcett.  An introduction to ROC analysis, 
Pattern Recognit. Lett. 27, (2006), pp. 861-874.  
(https://doi.org/10.1016/j.patrec.2005.10.010) 
[33] C.D Brown and H.T. Davis. Receiver operating 
characteristics curves and related decision 
measures: A tutorial. Chemom. And Intell. Lab. 
Syst. 80, (2006), pp. 24-38.  
(https://doi.org/10.1016/j.chemolab.2005.05.004) 
[34] E. K. Ampomah, Z. Qin, G. Nyame, F. E. Botchey, 
Stock market decision support modeling with tree-
based AdaBoost ensemble machine learning 
models, Informatica – An Int. J. of Comput. and 
Inform., Vol. 44,  No. 4 (2020).  
(https://doi.org/10.31449/inf.v44i4.3159) 
[35] R. Patel, S. Tanwani, C. Patida, Relation Extraction 
between Medical Entities using Deep Learning 
Approach, Informatica – An Int. J. of Comput. and 
Inform., Vol. 45 No. 3 (2021).  
(https://doi.org/10.31449/inf.v45i3.3056) 
[36] T. Czimmermann, G. Ciuti, M. Milazzo, M. 
Chiurazzi, S. Roccella, C.M. Oddo and P. Dario., 
Visual-Based Defect Detection and Classification 
Approaches for Industrial Applications—A 
SURVEY,  Sensors 20, Article No. 1459.  
(DOI:10.3390/s20051459) 
  

26 Informatica 46 (2022) 19–25 V. Asha