Zhongjian Li,

Jun Xiang,
Measuring the Geometrical Parameters
Lei Wang,
Ning Zhang,
of Slub Yarn Using a Computer Vision Based
Jing-an Wang,
Ruru Pan,
Image Sequencing Technique
Weidong Gao*
DOI: 10.5604/01.3001.0013.0739
Jiangnan University, Abstract
Ministry of Education, This article presents a computer vision method for measuring the geometrical parameters
School of Textiles and Clothing, of slub yarn based on yarn sequence images captured from a moving slub yarn. An image
Key Laboratory of Eco-textiles, segmentation method proposed by our earlier work was applied to segment sequence slub
Wuxi, Jiangsu, China yarn images to obtain overlapping diameter data. Then an image stitching method was
* e-mail: gaowd3@163.com proposed to remove the overlapped data based on the normalised cross correlation (NCC)
method. In order to detect the geometrical parameters of slub yarn, the frequency histogram,
curve fitting , and spectrogram methods were adopted to analyse the sequence diameter data
obtained. Four kinds of slub yarn with different geometrical parameters were tested using
the method proposed and Uster method. The experimental results show that the detection
results for slub amplitude, slub length, slub distance, and slub period obtained using the
method proposed were consistent with the set values and Uster results.

Key words: slub yarn sequence image, geometrical parameter measurement, image seg-
mentation, image stitching.

Introduction
Slub yarn is a simple fancy yarn whose
appearance is gained by the variation of
yarn linear density during the spinning
process [1]. It can be produced by mod-
ifying the ring spinning frame because
the intermittent acceleration of the roll-
ers causes varying degrees of draft [2].
This special appearance is determined by
the different geometrical parameters of
slub yarn, including slub amplitude, slub
length, slub distance, and the periodici-
ty rule of the slub. These parameters can
make woven fabrics have special aesthet-
ic properties [3, 4]. Therefore detecting
the geometrical parameters of slub yarn
is extremely significant for designing the
aesthetic properties of fabric [5].
The traditional method for analysing the
geometrical parameters of slub-yarn is
to count slubs in the yarn based on the
black boards, which demands special
experience from workers. However, this
method cannot get enough length of slub
yarn to detect the geometrical parameters
because the slub has no repeat in a short
length [6]. The method is also very tedi-
ous and time-consuming.
In recent years, many methods for
measuring slub geometrical parame-
ters automatically have been reported.
The principles of measuring slub ge-
ometrical parameters can be classified
into three main categories, which are
summarised with their pros and cons in
Table 1. These methods are as follows:
combining a capacitance-type sensor
with the voltage signals method [5, 7-9].
This method acquires mass data of slub
yarn using a capacitance-type sensor
first. And then the data are expressed as
voltage signals through a digital storage
oscilloscope or data acquisition card
(DAQ) etc. The Uster Tester is usually
utilised in this method. However, using
a capacitance-type sensor to obtain the
slub yarn diameter is an indirect meth-
od of detecting slub geometrical param-
eters. The detection result of slub yarn
shows a big difference from that with
the traditional visual method. The result
of this indirect method also depends on
the test environment conditions because
the capacitors are usually affected by
the temperature and humidity. The res-
olution of this method is relatively low,
where the capacitors sample data every
8 mm (e.g. old versions of the Uster test-
er) [10].
The Constant Tension Transport (CTT)
Tester method [11] was developed by the
Lawson and Hemphill Company. In the
tester, a Charged Coupled Device (CCD)
Table 1. Advantages and disadvantages of the methods for measuring the geometrical
parameters of slub yarn.
Methods Advantages
Capacitance-type – Primitive and dominant method
sensor with voltage – Fast
signals method
CTT Tester method – Does not depend on testing
environment
– High resolution
Image analysis method – Pixel-level resolution
– Not affected by yarn hairiness
– More accurate
array is used to collect yarn profile or
diameter data as the slub yarn moves
through the tester. And then an Oscillo-
scope and the threshold value are utilised
to transform the data into a voltage sig-
nal. Although this method is not affected
by the test humidity or fibre blend vari-
ation, it might be by the presence of lint
and yarn hairiness.
The second is the image analysis meth-
od [6, 12], where the slub yarn is usual-
ly wrapped on a black board. And then
a camera, flat scanner, etc. is adopted to
capture the slub yarn image. Next image
segmentation methods, which consist of
the threshold , Gabor Filter and morpho-
logical algorithm methods, are applied to
obtain the diameter data. Finally the data
are processed using the Histogram meth-
od, Cluster method, etc. However, a fast
device for obtaining a slub yarn image
continuously has not been presented thus
far. And an accurate image processing
method has also not beenproposed to de-
tect the geometrical parameters of slub
yarn from yarn sequence images.
Disadvantages
– Depends on the testing environment
– Indirect method
– Low resolution
– Affected by the presence of lint and
yarn hairiness
– Yarn under tension reduces the
diameter measured
– No detection instrument is formed

26 Li Z, Xiang J, Wang L, Zhang N, Wang J, Pan R, Gao W. Measuring the Geometrical Parameters of Slub Yarn Using a Computer Vision Based Image Sequencing Technique.
FIBRES & TEXTILES in Eastern Europe 2019; 27, 3(135): 26-35. DOI: 10.5604/01.3001.0013.0739
 For slub period measurement, spectrograms produced by the FFT method are adopted. A

flowchart of the inspection process is shown in Figure 1.

Segment image Remove hairiness

Subtract background Segment image IS Iog with K-means from Ib2 with two
image from image IS with small threshold clustering algorithm judgment templates
for image Ib1 for image Ib2 for image Icore

Figure 1. Flowchart of inspection process.

Above should be ‘Subtract backgroung image from..’ ‘segment image.. with small threshold’,
This work aims at tackling the prob-
lems of the image analysis method and
presents a computer vision method to
measure the geometrical parameters from
slub yarn sequence images. In our earlier
work [13], a novel method for measur-
ing the unevenness of the yarn apparent
diameter based on yarn sequence imag-
es had been proposed. In this method,
the K-means clustering algorithm was
employed to classify the pixels of each
frame in the sequence into two clusters –
yarn and background. Then the yarn core
was further extracted utilising the charac-
teristics of yarn hairiness from the classi-
fied image, and two judgment templates
were adopted to remove burrs, isolated
points and unrelated small areas in the
images. In that experiment, four kinds of
yarns were tested using this method and
an Uster Evenness Tester 5. The experi-
mental results showed that the method
proposed could accurately detect the un-
evenness of the yarn apparent diameter,
consisting of short-term, long-term, and
periodic variations.
In this study, an image acquisition device
was used to capture the overlapped slub
yarn sequence images. Then the segmen-
tation method proposed by our earlier
work was applied to segment the over-
lapped sequence images and an image
stitching method proposed to remove the
overlapping area based on the normalised
cross correlation method. Subsequently
sequence diameter data were obtained.
In order to calculate the slub amplitude
of each sample, a frequency histogram
of the sequence diameter data is first
constructed. And then the Gaussian dis-
tribution functions of each frequency
histogram are fitted. Finally the slub am-
plitude and reference level line value are
calculated using the coordinate values of
two wave crests and the cross point of
two fitting functions, respectively.
In order to measure the slub length and
slub distance, the sequence diameter data
are smoothed first. Then the smoothed
data are separated into 0 and 1 signals by
means of the reference level line value
obtained. Next the number of consecu-
tive 1 and 0 data is counted, and suspect
data are removed or combined by means
of two judging conditions. Finally the
slub length and slub distance are meas-
ured from the slub length histogram, slub

FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135) 27

 length of the
reported baseearlier
in our and SDi[13],
yarn, studies and BD are the diameter
sequence of the
yarn images areslub andinbase
taken yarn, respectively.
succession For
along the slub

andom
rn by anslub yarn, BL
area-scan i is arranged
CCD randomly.
camera with a resolution of 1024 × 768 pixels. A closed imaging box is used
the yarn, and to stabilise the yarn tension.
the image acquisition platform to
SL
shield from the disturbanceSLof2 ambient light. Meanwhile
Figure 3 shows the integrated structure
a special
1 of this system.

ht source is set up in the closedSD

box.
1
A yarn winding
BD mechanism,
SD2 servo motor, and
For tension
slub yarn samples #2, #3, and #4,
about a 100 m length of the slub yarn is
BL1 BL2 BL3 captured by the device above, respective-
ntroller are employed to draw the yarn, and to stabilise the yarn tension. Figure 3 shows thely.integrated For yarn sample #1, a 20 m length of
slub yarn is acquired because this sam-
Figure 2. Single period structure
Figure of slub-yarn.
2. Single period structure of slub-yarn. ple is not commercial yarn. In order to
ucture of this system.
obtain accurate sequence diameter data,
the neighboring slub yarn images should
Table 2. Four kinds of slub yarn samples. contain a certain level (e.g. 20-50%) of
overlapping area during acquisition. In
Parameters this experiment, a frame rate of 40 fps
Sample Base yarn was set to capture the overlapping im-
Slub Slub Slub Periodicity
number count/tex ages. The exposure time was uniformly
Length/cm Distance/cm Amplitude (%) Ruleset as 200 μs, and the speed of the image
acquisition device as 6 m/min. A back-
#1 27.8 5 5 200 Periodicground
slub image and four kinds of slub
yarn images with different yarn counts
were captured by the device. These im-
ages, cropped to a suitable width to allow
faster processing, are shown in Figure 4
with a size of 768×600 pixels. According
to the image calibration, a pixel in two
gure 3. YarnFigure
image3. acquisition
Yarn image acquisition
device: device: (1) closed
(1) closed imaging
imaging box,(2)
box, (2)area-scan
area-scan CCDCCD camera
cameradirections
sensor, (3)
occupies 5.21 μm in the image.
sensor, (3) yarn, (4) light source, (5) yarn guiding devices, (6) camera lens, (7) yarn tension
control panel, (8) touch screen, (9) output rollers with servo motor. The real length of yarn in this image is
approximately 4 mm.
rn, (4) light source, (5) yarn guiding devices, (6) camera lens, (7) yarn tension control panel, (8) touch
distance histogram, and Slub Length- amplitude, slub length, slub distance, and Image segmentation and stitching
Slub Distance distribution diagram. slub period. A single period structure of
reen, (9) output rollers with servo motor. In order to stitch the sequence slub yarn
slub-yarn is shown in Figure 2. In the
images, two adjacent yarn images of slub
For slub period measurement, spectro- figure, the slub yarn is composed of two
yarn #1, as shown in Figures 4.e1 and
grams produced by the FFT method are parts: the base yarn part and slub part; SLi
For slub adopted.
yarn samples #2, #3, and #4,, about a 100m length of the slub yarn
A flowchart of the inspection is the length of the slub, BLi the lengthis captured
4.e2, by
are the
used to explain the principle of
the image mosaic method. The specific
process is shown in Figure 1. of the base yarn, and SDi and BD are the
steps and results of this method are as
vice above, respectively. For yarn sample #1, a 20m diameter of the
length of slub
slub and
yarnbase
is yarn, re- because this
acquired follows:
spectively. For a random slub yarn, BLi is
Experimental details arranged randomly.
n Step 1: Remove the background infor-
mation. In this part, the background
mple is not commercial yarn. In order to obtain accurate sequence diameter data, the neighboring
Sample preparation slub
image Ibg is subtracted from all the
In the experiment, four kinds of 100% Sequence image acquisition original images Iog to eliminate the
cotton ring spun single slub yarns, as As reported in our earlier studies [13], impact
rn images should contain a certain level (e.g. 20–50%) of overlapping area during
shown in Table 2, were used for detecting sequence yarn images are taken in suc-
acquisition. In thisof non-uniform brightness.
The resulting images Is after the back-
the geometrical parameters. In the table, cession along the slub yarn by an ar- ground image Figure 4.e is subtracted
slub yarn sample #1, which is periodic ea-scan CCD camera with a resolution from Figures 4.e1 and 4.e2 are shown
slub yarn with a slub length and slub dis- of 1024 × 768 pixels. A closed imaging in Figures 5.a1 and 5.a2.
tance of 5cm, was specially made for this box is used as the image acquisition plat-
experiment. Sub yarn samples #2, #3, form to shield from the disturbance of n Step 2: Segment the resultant images
and #4, whose slub distance is random, ambient light. Meanwhile a special light Is and extract yarn core images. In this
were commercial slub yarn for producing source is set up in the closed box. A yarn part, small threshold values T1 can be
fabric. For a periodic slub yarn, the ge- winding mechanism, servo motor, and determined easily to segment image
ometrical parameters consist of the slub tension controller are employed to draw I to obtain a binary image I which
s b1
has clear hairiness. In order to obtain
an accurate yarn core image Icore, the
Table 2. Four kinds of slub yarn samples.
K-means clustering method is applied
Parameters to determine another threshold val-
Sample Base yarn
number count, tex Slub length, Slub distance, Slub amplitude, Periodicity ue to segment the original image Iog
cm cm % rule to obtain another binary image Ib2.
#1 27.8 5 5 200 Periodic slub And then two judgment templates
#2 27.8 4.5 13-20 150 Random slub [13], which are proposed by our ear-
#3 15.6 5 35-70 140 Random slub lier work, are applied to process the
#4 9.7 6 15-40 150 Random slub Ib2 images to obtain Icore images. By

28 FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135)

 768×600 pixels. According to the image calibration, a pixel in two directions occupies 5.21μm in
image. The real length of yarn in this image is approximately 4mm.

image. The real length of yarn in this image is approximately 4mm.

using all the Icore images, the diame-
ter pixel values of the slub yarn can a1) a2) b1) b2)
be calculated. The resultant images of
Ib1 and Icore are shown in Figures 5.a2
and 5.a3 and 5.b2-5.b3, respectively.
n Step 3: Eliminate the yarn core image
Icore from image Ib1 and execute dot
multiplication to obtain a test image
(TEST image). The resultant images
of subtracting image a3) from image c1) c2) d1) d2)
a2) and image b3) from image b2)
are shown in Figures 5.a4 and 5.b4.
TEST images of image a4, b4) dot
multiplication of the original images
(Figures 4.e1 and 4.e2) are shown in
Figures 5.a5 and 5.b5, respectively.
n Step 4: Image stitching. The normal-
ised cross correlation (NCC) method (a)
a)
[14-16] is applied to calculate the
stitch position between the TEST 1 e) e1)
(a) e2)
image and TEST 2 image in this pa-
per. The NCC method is a simple
template-matching method that deter-
mines the location of a desired pattern
represented by a template function, t,
inside a two-dimensional image func-
tion, f. A specific matrix T with a size
of H×W (W = 1, 2…N) in the TEST 2
image is selected as the calculated b)
template, as shown in Figure 6. H and
(b)
W are two adjustable parameters Figure 4. Images captured: (a) Four kinds of slub yarn images: a1) sample #1 – part image
which represent the height and width of of base yarn, a2) sample #1 – part image of slub, b1) sample #2 – part image of base yarn,
the template, respectively. Let T(x, y) b2)captured: sample #2 – part image of slub, c1) sample (b) #3 – part image of base yarn, c2) sample
Figure 4. Images #3 – part image (a)ofFour
slub, d1) kinds of slub
sample #4 – yarn images:
part image a1)yarn,
of base sample #1 -- part
d2) sample #4 –image
part of base y
be the intensity value of T at pixel (x, y), image of slub; (b) Sequence images of slub yarn sample #1, e) background image, e1) second
and f(x, y) be the intensity respectively.
Figure value of 4. fImages
at image, e2) first (a)
captured: image.
Four kinds of slub yarn images: a1) sample #1 -- part image of base y
pixel (x, y). f being the movement a2) sample ma-#1 -- part image of slub, b1) sample #2 -- part image of base yarn, b2) sample #2 --
trix in the TEST 1 image, whose size
a1)
is the same as T. The NCC a2) sample method is#1 -- part image of a2) slub, b1) sample a3)
#2 -- parta4)image of base a5)
yarn, b2) sample #2 --
evaluated at every point (u, v) for f image of slub, c1) sample #3 -- part image of base yarn, c2) sample #3 -- part image of slub, d1) sam
and T, which has been shifted over the
TEST 1 image by u-steps image in theofx-di- slub, c1) sample #3 -- part image of base yarn, c2) sample #3 -- part image of slub, d1) sam
rection and v-steps in the #4 y-direction.
-- part image of base yarn, d2) sample #4 -- part image of slub; (b) Sequence images of slub y
If the matrix f consists of just a part of
the intensity value of the #4TEST -- part 1 im- image of base yarn, d2) sample #4 -- part image of slub; (b) Sequence images of slub y
M MN age N (as shown by rectangles in Figure b1) b2) b3) b4) b5)
  T(Tx(, xy,)y)TTfu ,fv u,v f (fu(u x, xv,vM
6, the remaining  parts
y )yN ) 
in matrix f are
 
T ( x0.5, y )  T   f u,v  f (u  x, v  y) 
u  1, 2,,, M  H
u  1, 2,,, M  H  1  1
 (u, v) 
M N  (u , v )  M N
x 1x y1 1y 1

T ( x, y)  T  Tf u(,Tvx(,zero-filled. 1,2,,,  11(1) (1) u  1, 2,,, M  H  1
y )(All
u , v) the NCC coefficients
M MN N x 1 y 1 2 2 
0.5
v v  NN
1, 2,,, NN

M N
v  1, 2,,, N  N  1 (1)
2
f,)y(u)TTx,v2 
   xy      f
  (u
f (u x
,Mv
x ,vN y
 )y)
u  1, 2,,,f
 f 
u , v u
M 
H2  1 N
,v  M 2
0.5
1y 1
x1xy1 
1y 1 are stored x  y1in
1x 1y 1 a correlation
0.5   
T ( x, ymatrix ) T   γ(u,
(1)  f (u  x, v  y )  f u ,v  
M N 2 x 1 vy 1 1, 2,,, N  N  1x 1 y 1 
T ( x, y )  T     f v) (u as  y )  f u ,v in Equation (1) [17].
x, vdefined
2


sthe
themean
meanofofthe
thetemplate
templateandandf uf,vu ,that of f(x, y) in the region under the template. These
x 1
 y 1

v that of f(x, y) in the region under the template. These

Where, T is
Where, is the
themean
mean ofofthe
thetemplate
template and
and f u ,v that of f(x, y) in the region under the template. These
e template and f u ,v that f(x,y)y)in in
thatofoff(x, thethe region
region under under the
the template. These
:
as: template.
can be defined as:These can be defined as: Figure 5. Image processing results of two example images: a1)-b1) images after background
Figure 5. Imageremoval, processinga2)-b2)results of twowith
binary images example images:a3)-b3)
clear hairiness, a1)-b1) images
yarn after a4)-b4)
core images, background remo
11 H H WW
binary-hairiness images, a5)-b5) gray-hairiness images (TEST 1 and TEST 2 image).
TT
HH WWxx1 1y y1 1


TT( x(,xy, )y )   (2)
T
1 H W
(2)
(2)

T ( x, y ) (2)
1 H W
a2)-b2) binary (2) ximages with clear hairiness, a3)-b3) yarn core images, a4)-b4) binary-hairiness ima
H  W
T
H W
 T ( x, y ) 1 y 1
M N
 x 1 y 1
11 H H WW  T ( x, y)  T   fu ,v  f (u  x, v  y) 
f uf,vu ,v
H W HH 

WWx
f (fu(ux,xv, vy )y ) (3)
f 
1 H W


(3)
f ((uu, v
(3)
) x, v  y)x 1 y 1 (3)TEST 2 image).0.5
u  1, 2,,, M  H  1
(1)
1 1y 
x1  y1 1 a5)-b5)Hgray-hairiness
u , v
 W x 1 y 1 images (TEST 1 and v  1, 2,,, N  N  1 (1)
  
M N 2 M N 2
f u ,v  f (u  x , v  y ) (3)  T ( x, y )  T     f (u  x, v  y )  f u ,v  
   
H W Here,
x 1 y if1 u-x < 0 or v-y < 0 then f(u-x, x 1 y 1 x 1 y 1 
x<0ororv-y<0
<0 v-y<0then f(u-x,
then f(u-x, v-y)=0,
v-y)=0,andandf(0, 0)=TEST(1,1),1),f(0,
f(0,0)=TEST(1, f(0,1)=TEST(1,
1)=TEST(1,2),2),and andsosoon.on.
v-y) = 0,
Here, if u-x<0 and orf(0,0) = TEST(1,1),
v-y<0 then f(u-x,Step f(0,1)
v-y)=0, and f(0, 0)=TEST(1, 1), f(0, 1)=TEST(1, 2), and so on.
4: Image stitching. The normalised cross correlation (NCC) method[14-16] is applie
n f(u-x, v-y)=0,= TEST(1,
and f(0, 0)=TEST(1,2), and so1), on.f(0, 1)=TEST(1, Where, 2),Tand so on.
isEquation
the mean of the template and f u ,v that of f(x, y) in the region under the template. These
(1).

calculate as: position between the TEST 1 image and TEST 2 image in this paper. The N
can be definedstitch
FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135)the 29

method is a simple template-matching method

1 H W
that determines the location of a desired pat
 1
f u ,v   f (u  x, v  y )
H  W x 1 y 1
(3)

Here, if u-x<0 or v-y<0 then f(u-x, v-y)=0, and f(0, 0)=TEST(1, 1), f(0, 1)=TEST(1, 2), and so on.

Geometrical parameter
analysis and discussion
M N


M N


 TT (( xx,, yy))  TT   ff
x 1y 1
u ,v
u ,v  ff ((uu 
  xx,, vv 

 yy )) 
 1,
uu  2,,, M
1, 2,,, M H
H  11
 ((uu,, vv)) 
  x 1y 1
(1)
 MM NN 2 M N 0.5
2 0.5
 1,
vv  2,,, N
1, 2,,, NN
N 11 (1) Explanation of terms
 T ( x, y )  T   
 
M N
  2
T ( x , y )  T 2
 f (u  x , v  y )  f
  y 1
 f (u  x, v  y )  f uu ,,vv   Widthy of
 Template W
Reference level
x 1
y 1 x 1
x 1
y 1 x 1
y 1

The reference level is a line which is on

is
is the
the mean
mean ofof the template
the Heighy
template of and ff uu ,,vv that
and that of of f(x,
f(x, y) y) inin the
the region
region under
under thethe template.
template. These
These Overlapping
Region the level of the base yarn [18]. All slub
Template H
yarn calculations are based on this ref-
dd as:
as: erence level. Using this line, the slub
part can be separated from the slub yarn.
11 H
H W W
Thus determination of the reference lev-
RRand andMC,MC,and andaapanoramic
panoramic T
T  H  image 
image T
isis xx,, yy ))
(( obtained,
T obtained, as
asshown
shownininFigure Figure77b).
(2)
(2) b).InInthe thefigure
figurethe theredredregions
regions
el is very important for the inspection
H W W x
x 11 y
y 11
of the geometrical parameters of slub
present
resent the the width
width of
ffof MC.
MC.1 1 sequence image.yarn. In this paper, we use the Gaussi-
H W
Figure
Figure 6. 6.Computation
uu ,, vv 
Computation
 H  W  ff ((uu 
H W
principle
principle of NCC
 xx,, vv 
of NCC
 yy ))
methodmethod for slub yarnfor slub
(3) yarnimage.
sequence
(3) an distribution function of the slub yarn
H W  xx 11 yy 11
 
diameter data to confirm the reference
a) b)
In this paper, we set H=100 pixels and W=N. After the calculation, the correlation levelγ(u,
matrix value
v) automatically.
is As we know,
u-x<0
u-x<0 or or v-y<0 then f(u-x,
v-y<0 then f(u-x, v-y)=0, v-y)=0, andand f(0, f(0, 0)=TEST(1,
0)=TEST(1, 1), 1), f(0,
f(0, 1)=TEST(1,
1)=TEST(1, 2), 2), and
and so
so on.
on. there are more than two wave crests in
Correlation Coefficients

the frequency histogram of slub yarn di-

ained as shown in Figure 7 a). In the figure, there is a Heighy quiteof distinct peak at point N, which ameter denotes
data [6, 10]. Take Figure 8 as an
Template H
example, there are just two wave crests
Matching
position MR
in the frequency histogram, which means
t the position of point N is the optimum matching position of the two adjacent slub yarn images.
this slub yarn has just one type of slub. In
this frequency histogram, P1 and P2 are
Rows/pixels the frequency values of the first and sec-
ppose the position of point N is (up, vpColumns/pixels ), then the matching position in the TEST 1 image,ond that is, crest,
wave the respectively, and D and
1
D2 are the diameter values of the first and
D1 and D2FigureFigure 7.StitchStitch position ofinformation first andand mosaic result of two TEST twoimages.
TEST
TEST
rowFigure
the 7. 7. Stitchvalues
position
position information
information
MC, can and andcalculated
mosaic
mosaic result
result of
oftwo and images.
Dimages.
are diameter the second wave crest, respectively. second wave crest, respectively. Thus we
1 and D2 are the diameter values of the first and second wave crest
tching MR and matching column be as: respectively,
can use D1 to represent the diameter of
gure
gure 6.6. Computation
In thisprinciple
Computation paper, we
principle of
of NCCset H = 100
NCC method
method for pixels
for slub
sluband yarn
yarn sequence
The values
sequence image.
image. of MR and MC are adopted the base yarn part, and D2 to represent the
D1 to represent the diameter of the base yarn part, and D2 to represent the diameter Thus we of use D1 totwo
canoverlapping represent the diameter of the base yarn part, and D2 to represent th
When
When the the MR
MR After
W = N. valuesvalues theof of all all the
calculation, the sequence
sequence
the correla-
MR images
images
utop find
H are
are obtained,
obtained,
the overlapping the
the
line overlapping
between regions
regions
diameterare are
of the slub part. From the fre-
(4)
per,
per, we set H=100
we set tion pixels
H=100 matrixand
pixels γ(u,W=N.
and v) is After
W=N. obtained
After the
the calculation,
as shown inthe
calculation, adjacent yarn
the correlation
correlation matrix
matrix images
γ(u, v)inis
γ(u, v) isorder to stitch the quency histogram, we can also find a val-
om the frequency Figure histogram,
7.a. In the wefigure,can also there find is aa quite
valley between yarn images.the first wave
the slub
Finally crest
the part. Fromimage
original the frequency histogram, we can also find a valley between the fir
ley between the first wave crest and sec-
moved
movedand and distinct
sequence
sequence diameter
diameter data
data of ofthe theMCslub
slub
 yarn
yarn
vof  can
W
TESTcan1be be reordered.
reordered.
is separated basedThen
Then these
on MR theseandreordered
reordered data
data
own in Figure 7 a). In thepeak at
figure, point
there N,
is awhich
quite denotes
distinct
own in Figure 7 a). In the figure, there is a quite distinct peak at point N, which denotes peak atp point N, which denotes (5)ond wave crest, which can be determined
that be
crest, which can thedetermined
position ofbypoint fittingNthe isamplitude.
the opti- values.
histogram MC, and a panoramic
In order to and imagewave
second
determine is obtained,
crest, which by canfitting
be determined by fitting
the histogram the histogram
values. In ordervalues. In orde
eon
adopted
adopted mum matching position of the two ad- as shown in Figure 7.b. In the figure the to determine the reference level value,
on of pointto
of point to
N detect
N isdetect
is the the
the optimum
optimum thegeometrical
geometrical
matching
matching position parameters
parameters of
of the
the two of
ofthe
theslub
adjacent slubslub
yarn.
yarn. yarn
yarn images.
The values of
jacent MR slub andyarn MC images. areposition
adopted
Suppose3. Slubto
the
two
find
Length
adjacent
the
regions
(SL)
slub
overlapping
represent
and Slub
images.
the line
widthbetween
thereference
Distance MC. value,
of level
(SD) twothe adjacent
Gaussian
the yarn
distribution
Gaussian functions
distribution (F1 and(F1
functions F2) of all the w
el value, the Gaussian distribution functions (F1 and F2) of all the wave crests are
position of point N is (u , v ), then the and F2) of all the wave crests are fitted.
point N (upp,, vvpp), the TEST
TEST 11 image,
osition p p
osition of
of point N is
is (u ), then
then the
the matching
matching position position in in the image, thatthat is,
is, the
the all the sequence
matching position in the TEST 1 image, When the MR values fitted.ofThen a crossing point DThen a crossing
3 (as shown point8)Dbetween
in Figure 3 (as shown
D1 andinD2 can be calcu
ages
ssingin order
point Dthattois,stitch
3 (as shownthe matching
the yarnrow
in Figure images.
8) between
MR and WhenFinally
D1 and
match- thethe
D original
reference
can
2images beare image
level
calculated.
obtained, of TEST
is determined,
Finally
the 1the
overlappingis separated
slub
re- part based
can
Figure on
8bebetween
separated from
D1 and D2 the
can yarn.
be cal-As shown in
MR
MR and
and matching
matching column MC,
column MC, can be calculated as: gions are removed and sequence diame- culated. Finally the reference level value
ing column MC,can canbe becalculated
calculated as:as:
Geometrical
elGeometrical
value is obtainedparameter
parameter
by equation analysis
analysis
(6): and discussionter data of the slubtheyarn
anddiscussion reference level value is obtained by equation (6):
can be reordered. is obtained by Equation (6):
MR
MR  u
 u pp  H H     (4)
Figure 9, the ith slub
Then these(4) reordered (SL
length datai)are and ith to
adopted slub distance (SDi) can be calculated based on the
(4)
detect the geometrical parameters of the D  D3
1Explanation
Explanationofofterms terms MCD2v DW3     (5) slub yarn. D 2     (6) (
DMC  v p  W reference level from a(5) smoothed
(5)(6) diameter data signal. 2
p
2
Reference
Reference
es MR level
level
MC F1 function
es of MR and MC are adopted to find the
of and are adopted to find the overlapping
overlapping line
line between
between two
two adjacent
adjacent yarn
yarn P1
F1 function
P1
er to stitch the yarn images. Finally the original image of TEST 1 is separated based on
of TEST 1 is
eerreference
he to stitch thelevel
reference yarn
levelimages. Finally
isisaaline
linewhichthe original
which isison image
onthe
thelevel
level of
ofthe
the separated
base
base yarnbased
yarn [18].onAll
[18]. Allslub
slubyarn
yarncalculations
calculationsare
are Reference Level Value D
Frequency
Diameter value, pixels

Reference Level Value D P2

Frequency

F2 function
sed
ed on
onthis
thisreference
referencelevel.
level.Using
Using
P2this
thisline,
line,the
theslub
slubpart
partcan
canbe
beseparated
separatedfrom
fromthe
theslub
slubyarn.
yarn.Thus
Thus
F2 function

ermination
terminationof
ofthe
thereference
referencelevel
levelisisvery
veryimportant
importantfor
forthe
theinspection
inspectionof
ofthe
thegeometrical
geometricalparameters
parameters
D1 D3 D D2
Slub yarn diameter
slub
slubyarn.
yarn.InInthis
thispaper,
paper,we
weuse
usethe
DD
theGaussian
Gaussiandistribution
distributionfunction
functionof
ofthe
theslub
slubyarn
yarndiameter
diameterdata
datatoto
D1 D3 2
Slub yarn diameter
Figure 8. Frequency histogram of
Distance, cm
slub yarn diameter data.
nfirm
nfirmthe
thereference
referencelevel
levelvalue
valueautomatically.
automatically.As
Aswe
weknow,
know,there
thereare
aremore
morethan
thantwo
twowave
wavecrests
crestsinin
Figure 8. Figure
Frequency histogram
8. Frequency of slubofyarn
histogram diameterFigure
slub yarn 2. Slub
data. 9. Smoothed Amplitude
diameter data signal.
diameter data. Figure 9.
efrequency
frequencyhistogram
histogramof
of slub
slubyarn
yarndiameter
diameterdata
data[6,
[6,10].
10].Take
TakeFigure
Figure asSmoothed
88as an diameter
anexample,
example,there data
are signal.
thereare
e In this paper, the slub amplitude (SA) is defined as:
30 FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135)
st
t two
twowave
wavecrests
crestsininthe
thefrequency 4. Slub Period
frequencyhistogram,
histogram, which
whichmeans
meansthis
thisslub
slubyarn
yarnhas
hasjust
justone
onetype
typeof
ofslub.
slub.
Di 1
slub amplitude (SA) is defined as: SAi  100% i 
1, 2,..., N (
D1
 of functions F1 and F2. After the function approximation and cross point calculation, the D1, D2, D3,

D1 and D2 are the diameter values of the first and second wave crest, respectively.
D, and SA values of the four kinds of slub yarn are listed, shown in Table 3.
amplitude.
D1 to represent the diameter of the base yarn part, and D2 to represent the diameter of
b)
om the frequency histogram,a)we can also find
3. Slub Lengthbetween
a valley (SL) andthe
Slub
firstDistance (SD)
wave crest

When
crest, which can be determined by fitting the the reference
histogram values. level is determined,
In order to determinethe slub part can be separated from the yarn. As shown in

Frequency
Frequency

Figure
el value, the Gaussian distribution functions (F1 and F2)ith
9, the of slub length
all the wave(SL are ith slub distance (SDi) can be calculated based on the
i) and
crests

reference
ssing point D3 (as shown in Figure 8) between level
D1 and D2from
can abesmoothed diameter
calculated. Finallydata signal.

el value is obtained by equation (6):

D2  D3
D Slub Yarn Diameter Value/pixels (6) Slub Yarn Diameter Value/pixels
2
c) d)
F1 function
P1

Reference Level Value D

Frequency

Frequency

Frequency
P2
F2 function

Figure 9. Smoothed diameter data signal.

D1 D3 D D2 4. Slub Period
Slub yarn diameter Slub Yarn Diameter Value/pixels Slub Yarn Diameter Value/pixels

The spectrogram is an ideal tool to check the repeat pattern of slubs. In order to obtain the slub
Figure 10.histogram
Figure 8. Frequency Frequencyofhistograms
slub yarnand Gaussian
diameter data.distribution functions of four kinds of slub yarn.
Figure 10. Frequency histograms and Gaussian distribution functions of four kinds of slub yarn.
period, the Fast Fourier Transform (FFT) is used to analyse the spectrogram base for 1D continuous
e
Slub amplitude slub yarn diameter data yarnindiameter
this paper. data in this paper.
Wavelengths (Wl) Wave- second
are calculated basedwave crest,
on the of the D3 rep-
respectively.
resolution
In this paper, the slub amplitude (SA) is lengths (W ) are calculated based on the resents the diameter
slub amplitude (SA) is defined as:
defined as:
Table 3. Five numerical resolution
l
of
results
the slub
ofimage,
yarn
four askinds in
of slub
point of
yarn.F1 andofF2.theAfter
functions
value crossing
the
slub yarn image, as in equation (8).
Equation (8). function approximation and cross point
Di 1
SAi  Slub
D1 yarni 
100% 1, 2,..., N  (7) (7) N calculation, the D1, D2, D3, D, and SA
D1/pixels  Wl D2/pixels i
D
N i  [0,1,..., N
3/pixels
t ]  (8) values D/pixels
of the four kinds of
(8) SA slub(%)
yarn are
Where, SAisample
represents the ith slub ampli- Nt  t listed, shown in Table 3.
ith slub
represents thetude, andamplitude, and Di+1 value
Di+1 the diameter the diameter
of the value of theNti+1th
Where, is thewave total length of the 1D
i+1 th wave crest #1 in the frequency Where, 49
Nt is thecontinuous
histo- 97
total lengthslub
of theyarn
1D continuous
diameter data, 65 yarn
slub Δt diameter
It can be 81
data, Δt from
seen Table 3198
the resolution of the D1
that
gram.
uency histogram. D1the
D1 is is the diameter
diameter value
value of the
of the firstfirst the resolution
wave crest of the slub yarn image, and values of the four samples are #4<#3<#2
in the frequency
wave crest in the #2frequencythehistogram.
slub yarn45 If Δt
image, = Δt
and 5.21 μm62
=5.21μm in in
thisthis
paper.
paper. 55 & #1. Those 59 of #1 and #2 138 are simi-
lar, which is consistent with the actu-
re are just twothere
wavearecrests #3
justintwo
the wave crests
frequency in the
histogram, 29
fre-
there is only one type39 of slub 37 38 134
quency histogram, there is only one type Experimental results and discussion al base yarn count of the four samples:
of slub amplitude.#4 21 Frequency histogram 29 26 #4<#3<#2=#1. 28 In the Table, 138the reference
In order to determine the reference level line values D of the four samples deter-
3.2 Experimental results and discussion
Slub length (SL) and slub distance value D and slub amplitude SA, the fre- mined are 81, 59, 38, and 29 pixels. It
It can
(SD) be seen from the table above quency that
histogramsthe ofDfour1 values
kinds ofof slubthe canfouralso samples
be found thatare the #4<#3<#2
SA values of & #1.
When the reference level is determined, yarn are shown in Figure 10. In these fig- the four samples (198%, 138%, 134%,
the slub part can be separated from the ures, the Gaussian distribution functions 138%) are close to the set values (200%,
Those of #1
yarn. Asand
shown #2 in are similar,
Figure slub F1isand
9, the ith which F2 of each slub yarn are fitted 150%, 140%, 150%). Thus the method
consistent with the actual base yarn count of the four samples:
length (SLi) and ith slub distance (SDi) can based on the histograms. D1 and D2 rep- proposed can be applied to inspect the
be calculated based on the reference level resent the diameter values of the first and slub amplitude.
from a smoothed diameter data signal.
Table 3. Five numerical results of four kinds of slub yarn.
Slub period
The spectrogram is an ideal tool to check Slub yarn sample D1/pixels D2/pixels D3/pixels D/pixels SA, %
#1 49 97 65 81 198
the repeat pattern of slubs. In order to
#2 45 62 55 59 138
obtain the slub period, the Fast Fourier
#3 29 39 37 38 134
Transform (FFT) is used to analyse the
#4 21 29 26 28 138
spectrogram base for 1D continuous slub

FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135) 31

 Through the three steps above, the resultant data of the four samples are shown in Figures 11

d4). From these figures, we can see that the outliers in the blue dotted rectangles are almost (b) sample #2
(a) sample #1
oved or combined.

Diameter value,
Diameter value,

pixels
pixels

Distance, cm Distance, cm

Diameter value,
Diameter value,

pixels
pixels

Distance, cm Distance, cm

0 1 Signal
0 1 Signal

Distance, cm Distance, cm

0 1 Signal
0 1 Signal

Distance, cm Distance, cm

(b) sample#1
a) sample #2 (c)b)sample
sample#3
#2
(a) sample #1
Diameter value,

Diameter value,
pixels

pixels

Distance, cm Distance, cm
Diameter value,

Diameter value,
pixels

pixels

Distance, cm Distance, cm
0 1 Signal

0 1 Signal

Distance, cm Distance, cm
0 1 Signal
0 1 Signal

Distance, cm Distance, cm

c) sample #3 d) sample #4
(c) sample #3 (d) sample #4

Figure 11. Part diameter data of four samples: a1)-a4) Original sequence diameter data, b1)-b4) Smoothed diameter data, c1)-c4) Original
“0 and 1” data signals, d1)-d4) The processed “0Figure
and 1”11.
dataPart diameter data of four samples: a1)-a4) Original sequence diameter data, b1)-b4)
signals.

Slub length and slub distance analysis
Since irregular variation of slub yarn di-
ameter data exists, there are some burrs
and noises in the original sequence di-
ameter data. These may cause some er-
rors when calculating the slub length and
slub distance using the reference level
value D. Therefore the diameter data
need to be smoothed first. A part of the
original diameter data and the smoothed
data of the four yarn samples are shown
in Figures 11.a1-11.a4 and 11.b1-11.b4.
After the diameter data is smoothed, the
smoothed data which is greater than the
reference level value D is set 1 and the
other – set 0. The “0 and 1” data signals are
shown in Figures 11.c1-11.c4.(d) sample #4
From Fig-
Smoothed diameter data, c1)-c4) Original ―0 and 1‖ data signals, d1)-d4) The processed ―0 and 1‖
ures 11.c1-11.c4, we can see that although
the original data have been smoothed first,
there are also some outliers, as shown in
the dot blue rectangle. In order to obtain
accurate results, the following steps are
applied to process the “0 and 1” data sig-
nals in Figures 11.c1-11.c4:
n Step 1: Count the number of consec-
utive 1 and 0 data from the “0 and
1” data signal, and denote them as
N1, i (representing the slub length se-
quence, i = 1,…, n) and N0, i (repre-
senting the slub distance sequence,
i = 1,…, n), as shown in Figure 11.c.
n Step 2: The slub length and slub
distance will be not be shorter than
20 millimeters during the yarn pro-
duction [6]. Therefore if N1, i < 2 cm,
N0, i <2 cm, and N0, i +1 < 2 cm, the
data in N1, i will be considered as the
base yarn part and set as 0.
n Step 3: If N1, i > 2 cm, N0, i < 2 cm, and
N0, i  +1 < 2 cm, the data in N1, i and N1, i+1
will be combined together and set as 1.
Through the three steps above, the result-
ant data of the four samples are shown
in Figures 11.d1-11.d4. From these Fig-
ures, we can see that the outliers in the
dotted rectangles are almost removed or
combined.
When the processed “0 and 1” data sig-
nals are obtained, the slub length and

re 11. Part diameter

32 data of four samples: a1)-a4) Original sequence diameter data, b1)-b4) FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135)

othed diameter data, c1)-c4) Original ―0 and 1‖ data signals, d1)-d4) The processed ―0 and 1‖
distribution diagram of the four samples are constructed, shown in Figure 12.
counted. Then a slub length histogram, slub distance histogram and Slub Length-Slub Distance
distribution diagram of the four samples are constructed, shown in Figure 12.

distribution diagram of the four samples are constructed, shown in Figure 12.

a1) b1) c1)

Slub Distance/cm
Frequency

Frequency
(a) sample #1
(a) sample #1
Slub Length/cm Slub Length/cm Slub Length/cm
(a) sample #1
a) sample #1

(a) sample #1
a2) b2) c2)

Slub Distance/cm
Frequency

Frequency

(b) sample #2
Slub Length/cm Slub Length/cm Slub Length/cm

(b)b)sample
sample #2 #2

(b) sample #2
a3) b3) c3)

Slub Distance/cm
(b) sample #2
Frequency

Frequency

Slub Length/cm
(c) Slub
sample
Length/cm
#3 Slub Length/cm
c) sample #3
(c) sample #3
a4) b4)
(c) sample #3 c4)
Slub Distance/cm
Frequency

Frequency

(c) sample #3

Slub Length/cm Slub Length/cm Slub Length/cm

d) sample #4

(d sample)
Figure 12. Detection results of four slub yarn samples: a1)-a4) #4
frequency histogram of slub length, b1)-b4) frequency histogram of slub
distance, c1)-c4) distribution diagram of slub length-slub distance.
(d sample) #4
Figureslub12.distance
Detection results of four slub yarn
can be counted. Then ure 10). Thus
samples：a1)-a4)
(d sample) #4 that the frequency
we can determine
histogram of slub length.
It can be seen from the table above that
Figurea slub length histogram, slub distance four samples have just one type of slub the slub length and slub distance meas-
12. Detection results of four slub yarn samples：a1)-a4) frequency histogram of slub length.
histogram and Slub Length-Slub Dis- length. The abscissa
b1)-b4) (d sample)value#4
of each crest ured are very close to the set values.
12.frequency
Figuretance Detection
distribution histogram
results
diagram of the of
offour slub
four slubdistance,
wave inyarn
c1)-c4) distribution
samples：a1)-a4)
Figures 12.a1-12.a4 is meas- frequency
This candiagram that of
the slub
histogram
prove length-slub
of slub
reference length.
level
samples are constructed, shown in Fig- ured as the slub length of each sample, as value we selected from the histogram
b1)-b4) frequency histogram of slub distance,
ure 12. c1)-c4) distribution diagram
is correct. From the of slub
Table, length-slub
we can also
Figure
distance.12. Detection results of four shown slub inyarn
Table samples：a1)-a4)
3. frequency histogram
find that sample of one
#1, which has slub length.
type
b1)-b4) frequency histogram
From Figures 12.a1-12.a4, we can find of slub distance, c1)-c4) distribution diagram of slub length-slub
of slub length and slub distance, is peri-
It can be seen from Figures 12.b1-12.b4
distance.
that although the slub length is dispersed odic slub yarn. The slub period of sam-
and Figures 12.c1-12.c4 that there are ple #1 is about 10 cm. For samples #2,
b1)-b4) frequency
in different intervals, histogram of slub
there is no obvious distance, c1)-c4) distribution diagram of slub length-slub
distance.
multimodal on these histograms. Further- one, three, six, and five dominant peaks. #3, and #4, they all have just one type
more the four samples all have two wave Thus the slub distance can be measured of slub length and different types of slub
crests in the frequency histogram of the using these peaks. The results measured distance. Thus the three samples are all
distance.
sequence diameter data (as shown in Fig- are also listed in Table 4. random slub yarn.

FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135) 33

 of samples #2, #3 and #4 obtained from the Uster Tester are shown in Figures 13 a1), b1), and c1),
spectively. The FFT spectrograms of samples #2, #3 and #4 produced from the slub yarn sequence
respectively. The FFT spectrograms of samples #2, #3 and #4 produced from the slub yarn sequence
ages are also illustrated in Figures 13 a2), b2), and c2).
images are also illustrated in Figures 13 a2), b2), and c2). be attributed to mechanical faults (about
3.32 cm) plus the inter-slub distance
(4.98 cm). The dominant peak at 9.97 cm
can be attributed to the pattern of the long
slub (4.95 cm) plus its inter-slub distance
Amplitude

(4.98 cm). Thus we can conclude that

the slub period of sample #1 is 9.97 cm,
which is very close to the set value of
10 cm.

From Figures 13.b, we can find that the

Log Distance/cm dominant peaks on spectrograms ob-
(a) sample #1 tained from FFT are very consistent with
a) sample #1 the Uster spectrograms. The position of
(a) sample #1 dominant peaks in the Uster (located at
point ‘P’, ‘A’ or ‘B’) and FFT spectro-
grams (located at point ‘P’’, ‘A’’ or ‘B’’)
are all similar for samples #2, #3 and #4.
The corresponding wavelengths range of
point ‘P’, ‘A’ and ‘B’ in the Uster spec-
Amplitude

Amplitude

Amplitude

trograms and the corresponding wave-

lengths of point ‘P’’, ‘A’’ and ‘B’’ in the
Log Distance/cm Log Distance/cm Log Distance/cm
FFT spectrograms for samples #2, #3,
b) samples #2, #3 & #4 and #4 are listed in Table 5.
(b) samples #2, #3 & #4
Figure 13. Spectrograms for four yarn samples: (a) FFT, #1; a1) Uster, #2; b1) Uster, #3; From Table 5, we can conclude that the
Figure c1)
13. Uster,
Spectrograms for four
#4; a2) FFT, #2; yarn samples:
b2) FFT, (a)FFT,
#3; c2) #1; a1) Uster, #2; b1) Uster, #3; c1) Uster,method proposed for measuring the pe-
FFT,#4.
riodicity of slub yarn is quite consistent
(b) samples #2, #3 & #4 with the Uster method. Especially for
#4; a2) FFT, #2; b2) FFT, #3; c2) FFT, #4.
Spectrogram yarn sequence images are also illustrated the ‘P’ and ‘P’’ points, the two kinds of
The spectrogram is utilised to detect the in Figures 13.a2-13.c2. spectrograms have almost the same val-
gure 13. Spectrograms
From 13 for
(a), four
Figurevariations weofcanyarn samples:
seeyarn.
that there (a) FFT, #1; a1) Uster, #2; b1) Uster, #3; c1) Uster,
periodic slub Due are
to four peaks on the spectrogram produced by the FFT ues. However, there are also some incon-
sample #1 being produced specifically in From Figure 13.a, we can see that there sistencies between the two spectrograms
; a2) FFT,
method#2;
forb2)slub
a shorterFFT, #3;
yarn
length, c2)
sample
the FFT,
Uster #4. was
#1,Tester
which are located
not are at about 3.32, 4.98,
four peaks on the8.18 and 9.97 cm.
spectrogram The peak
pro- with respect to the three samples because
used to detect its spectrogram. The spec- duced by the FFT method for slub yarn of the principle of the Uster spectrogram
trogram of around
at the wavelength sample3.32cm
#1 generated by the tosample
can be attributed #1, which
mechanical faults are
and located
drafting at and FFT spectrogram. The Uster spectro-
aboutand that
waves,
From Figure 13 (a), we can see that there are four peaks on the spectrogram produced by the FFT
FFT method is shown in Figure 13.a. 3.32, 4.98, 8.18 and 9.97 cm. The peak gram divides the interval of the channel
locatedThe spectrograms
around 4.98cm can of be
samples #2, #3
attributed andinter-slub
to the at the distance.
wavelength
The around cm can canand
3.32 8.18cm
peak around be
calculates the integral of the channel,
whereas the FFT spectrogram dies not.
ethod for slub yarn sample #1, which are located at about 3.32, 4.98, 8.18 and 9.97 cm. The peak
#4 obtained from the Uster Tester are be attributed to mechanical faults and
shown in Figures 13.a1-13.c1, respec- drafting waves, and that located around
Due to samples #2 #3, and #4 being ran-
tively. The FFT spectrograms of samples 4.98 cm can be attributed to the inter-slub
the wavelength around domand
slubthat
yarn and that irregular variation
#2, #3 and #43.32cm canfrom
produced be attributed
the slub to mechanical faults and drafting
distance. The peak around 8.18 cm can waves,
exists, it is hard to obtain the arrangement
of different slub distances from the spec-
cated aroundTable
4.98cm can be attributed to the inter-slub distance.
4. Detection results of slub length and slub distance.
The peak around 8.18cm can above.
trograms be However, for sample #1
the arrangement of different slub distanc-
Slub length Slub length Slub distance Slub distance es can be obtained from the spectrogram
Yarn sample
measured, cm set, cm measured, cm set, cm because sample #1 is not random slub
#1 4.95 5 5.1 5 yarn. Finally the geometrical parameters,
#2 4.78 4.5 13.2, 16.2, 19.2 13-20 consisting of the slub amplitude, slub
37.5, 44.7, 51.3 length, slub distance, and the slub period,
#3 5.66 5 35-70
54.6, 60.6, 67.8
of the four samples all are detected. And
15.2, 20.8, 30.4,
#4 5.84 6
36.7, 41.2
15-40 the results analysed are in high accord-
ance with the set values and Uster Tester
Table 5. Corresponding wavelengths of the dominant peaks on two kinds of spectrograms. results.

Dominant peak points #2 #3 #4

P [19 cm, 21 cm] [50 cm, 60 cm] [20 cm, 21 cm] Conclusions
P' 21.3 cm 57.7 cm 21.9 cm In this paper, the geometrical parameters
A [7.5 cm , 11 cm] [14 cm, 15 cm] [41 cm, 43 cm] of slub yarn are measured from slub yarn
A' 11.2 cm 19.3 cm 41.5 cm
sequence images using image process-
B – – [11 cm, 13 cm]
ing technology. The image segmentation
B' – – 12.9 cm
method proposed in our earlier work is

34 FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135)

applied to obtain the overlapping diam-
eter data of slub yarn. Then an image
stitching method is utilised to remove the
overlapping part based on the NCC meth-
od to gain the sequence diameter data of
the slub yarn. Finally the sequence diam-
eter data are used to detect the geometri-
cal parameters of the slub yarn.
In the experiment, the reference level
value and slub amplitude of each sample
were determined first from the frequen-
cy histogram of the sequence diameter
data. Then the slub length and slub dis-
tance were measured by the slub length
histogram, slub distance histogram, and
Slub Length-Slub Distance distribution
diagram. Finally the slub period was ana-
lysed by means of the spectrogram.
From the measurement of the slub ampli-
tude, the detection results of four samples
are as follows: 198%, 138%, 134%, and
138%. They are close to the set values of
200%, 150%, 140%, and 150%.
For the measurement of slub length, the
detection results of four samples are:
4.95, 4.78, 5.66, and 5.84 cm. They are
also very close to the set values as 5, 4.5,
5, and 6 cm.
From the measurement of the slub dis-
tance, we detect that sample #1 has just
one type of slub distance, with a value of
5.1 cm, which is close to the set value of
5 cm. For samples #2, #3 and #4, there are
three, six, and five types of slub distance,
respectively, whose values are (13.2 cm,
16.2 cm, 19.2 cm), (37.5 cm, 44.7 cm,
51.3 cm, 54.6 cm, 60.6 cm, 67.8 cm),
and (15.2 cm, 20.8 cm, 30.4 cm, 36.7 cm,
41.2 cm). They are all similar to the set-
ting interval, namely [13 cm, 20 cm],
[35 cm, 70 cm], and [15 cm, 40 cm].
From the spectrograms of the four sam-
ples, we can conclude that sample #1 is
periodic slub yarn, with a slub period
9.97 cm, which is constant with the set
FIBRES & TEXTILES in Eastern Europe 2019, Vol. 27, 3(135)
value of 10 cm. For samples #2, #3 and
#4, spectrograms detected all conform
to the Uster spectrograms. Finally, com-
bining the measurement results of slub
length and slub distance with the spectro-
grams of the three samples, we can con-
clude that samples #2, #3, and #4 are all
random slub yarns. Therefore the method
proposed can be used to measure the ge-
ometrical parameters of slub yarn, and it
can obtain accurate results.
Acknowledgements
The authors would like to acknowledge the
Research Innovation Program for College
Graduates of Jiangsu Province (Grant No.
KYLX_1132); the National Natural Sci-
ence Foundation of China (grant number
61802152), the Natural Science Founda-
tion of Jiangsu Province (grant number
BK20180602), the China Postdoctoral Sci-
ence Foundation Funded Project (grant
number 2018M640453), the Jiangsu Prov-
ince Postdoctoral Science Foundation (grant
number 2018K037B), and the Fundamental
Research Funds for the Central Universities
(grant number JUSRP51907A).
References
 1. Gong RH, Wright RM. Fancy Yarns:
Their Manufacture and Application.
Woodhead: Crc Press; 2002.
 2. Wang J, Huang X. Parameters of Rotor
Spun Slub Yarn. Textile Research Jour-
nal 2002; 72:12-16.
 3. Hermanne L and Quintelier G. Law of
Critical Yarn Diameter and Twist: Influ-
ence on Yarn Characteristics. Textile
Research Journal 1951; 21:166-168.
 4. Grabowska KE. Characteristics of slub
fancy yarn. FIBRES & TEXTILES in
Eastern Europe 2001; 9: 28-30.
 5. Liu J, Xie Z, Gao W, et al. Automatic
Determination of Slub Yarn Geometrical
Parameters Based on an Amended Simi-
larity-based Clustering Method. Textile
Research Journal 2010; 80: 1075-1082.
 6. Pan R, Gao W, Liu J, et al. Recognition
the Parameters of Slub-yarn Based on
Image Analysis. Equipment Manufactur-
ing Technology 2011; 6: 25-30.
 7. Liu J, Li Z, Lu Y, et al. Visualisation
and Determination of the Geometrical
Parameters of Slub Yarn. FIBRES &
TEXTILES in Eastern Europe 2010; 78:
31-35.
 8. Elkhalek RA, Elbealy R, and Eldeeb A.
A Computer-Based System for Evalua-
tion of Slub Yarn Characteristics. Jour-
nal of Textiles 2014; 2014: 1-11.
 9. Lv H, Ma C. Using Capacitance Sensor
to Identify the Appearance Parameters
of Slub Yarn. Heidelberg: Advanced Re-
search on Electronic Commerce. Web
Application, and Communication; 2011.
10. Eldessouki M, Ibrahim S, Militky J.
A dynamic and robust image processing
based method for measuring the yarn
diameter and its variation. Textile Re-
search Journal 2014; 84: 1948-1960.
11. Sudhakar J. Characterization methods
and physical properties of Novelty yarn,
Textile Management and Technology
2005; pp. 130-134.
12. Liu X, Su Z, Wen Z, et al. Slub Extraction
in Woven Fabric Images Using Gabor
Filters. Textile Research Journal 2008;
78: 320-325.
13. Li Z, Pan R, Zhang J, et al. Measuring
the unevenness of yarn apparent diame-
ter from yarn sequence images. Mea-
surement Science & Technology 2016;
27: 015404.
14. Chien Y. Pattern classification and
scene analysis. IEEE Transactions on
Automatic Control 2003; 19: 462-463.
15. Lewis JP. Fast Normalized Cross-Cor-
relation. Circuits Systems & Signal Pro-
cessing 1995; 82: 144-156.
16. Hii AJH, Hann CE, Chase JG, et al. Fast
normalized cross correlation for motion
tracking using basis functions. Comput-
er Methods & Programs in Biomedicine
2006; 82: 144-156.
17. Russ JC. Template matching and corre-
lation, in: The Image Processing Hand-
book, 2 ed. Raleigh: CRC Press; 1994.
18. Uster tester 5-S8000 application report
measurement of slub yarns part1-basics
[homepage on the Internet]. c2007 [up-
dated 2007 Jun; cited 2018 Jan]. Avail�-
able from https://www.uster.com/en/
knowledge/textile-know-how/yarn-test-
ing/.
Received 24.01.2018 Reviewed 28.11.2018
35