You are on page 1of 4

Determination of Weave Type in Woven Fabric by Digital Image Processing

By Mizuho Kinoshita, Yositada Hashimoto, Ryuichi Akiyama and Sei Uchiyama, Members, TMSJ

Faculty of Textile Science, Kyoto Institute of Technology, Matsugasaki, Sakyo-ku, Kyoto 606

BasedontheJournalof theTextileMachinery
SocietyofJapan,Proceedings,
Vol.40,No.l 1P463-P467
(1987-11)

Abstract
Weavetypesofcommercial wovenfabricsweredetermined byanalysing thetwo-dimensional
Fouriertransfor-
mationpatternsof thereflectedimages.Datacollection anddataanalysisweredonebyimageprocessing system
basedon a microcomputer. Therelationshipsbetweenthestructural unitsof weavetypesandtheFourierpower
spectraaresummarized as follows:(1)structuralunitis observedintheunitcelldefinedbytheneighboring two
warpyarnsandtwofillingyarnsinFourierspace,(2)weaveharrnessisobtainedbycountingthenumberof power
spectrainsidethe unitcell,(3) structuralunitis equivalent
to the Fourierpowerspectrumarrangement of the
reflectedimagerotatedby90degree.

also been done using a microcomputer181. But there are very few
1. Introduction investigations on the pattern recognition of the woven fabrics
with regard to the fabric analysis. Although the structural
A woven fabric made of the cross combination of the warp properties of the fabrics are directly observed by visual meas-
and filling yarns has a two-dimensional lattice structure. To urements, it is useful to study the experimental methods of
study the fabric structure, it is useful to observe the periodic detecting the characteristic patterns automatically by using the
arrangement of the surface contours associated with the yam digital and/or optical image processing. We have already
faces or the geometrical intersticies enclosed by neighboring described the details on the analytical procedures for detecting
two yams. They have an effect on the optical and mechanical weave type and experimental demonstration by observing the
properties of the fabric. In practice, the quality of the woven diffraction pattems~9l. In this paper, we will demonstrate the
product is related tothe geometricalshapeof the lattice structure above analytical procedures by using the ordinary digital image
and their irregularitydistribution, and the visual and handling processing.
properties are also related to the surfacecontours of the fabric.
The fabric analysis is made to determine the fabric structure. 2. Experimetal Procedures
The real time data processingand the automatical method for
the detection and discriminationare required in place of the The reflected and transmitted images of the woven fabrics
manual measurement. were obtained by using the commercial television camera as
We have mainly studied an optical image evaluation on the shown in Fig. 1. A light source is a tungusten lamp. The
periodic arrangement of the fabric structure by observing the magnification of the image was made by the concave lens placed
optical diffraction pattern. In the previous papers, we have in front of the television camera. The light was incident upon
described the detection of fabric defects l'21,the discrimination an angle of with the normal to surface of the sample.
among fabric defects131, and the local irregularityunderuniaxial An image processing system (see Fig. 2) was used in this
extension~41 experiment was constructed by means of a microcomputer as a
In these above analysis, the two-dimensional arrangement base. Images were so magnified that the number of the yams
defined by yam intersticieshavebeen investigated.The surface could be 20-30 ends or picks on the screen. Each image was
contour patternis also one of the periodic arrangementdefined digitized into 128x128 pixels. Each pixel was assigned a gray
by the structralunit arrangement,i.e. weave type. level from dark to light of 0-128. This instrument allows a gray
The weave type of a woven fabric is constructed by the
repetition in space of a structuralunit defined by the warp and
filling faces. The theoretical studies of the construction for the
complete design have been made by several investigators~s-71
Numerical method for constructing the complete weaves has

Fig. 1 Geometrical arrangement to obtain reflected images of woven


Fig. 2 Schematic diagram of image processing system.
fabrics.

Vol. 35, No. 2 (1989) 1


level range to be pre-set, so that only those features of interest
are measured. Television camera was set at 0° overtically to the
sample.

3. Experimental Results and Discussion

The weave types are mainly classified into the following four
types: plain, twill, satin and derivatives. Figure 3 shows the
fabric structure models of the regular weaves on the upper side
and the surface contour pattern of the warp face models on the
lower side. In order to determine the weave type of the fabric
experimentally, it requires to obtain the surface contour pattern
of the woven fabric and the structural units associated with the
Fig. 3 Fabric structure models made of warps and fillings; (a) plain, weave types as indicated by dashed lines in Fig. 3(c), (d) and
(b) twill and (c) satin weave. Warp surface contours of plain (d), (e). There are two questions to answer: (1) How can we obtain
twill (e) and satin (f). the surface pattern associated with the warp faces or filling
faces? (2) How can we detect the structural unit of the weave
type in the surface contour pattern?
According to the photometric measurements of the light
intensity reflected from the fabric, the intensity distribution
depends on the angle between the incident light and the reflected
light, and the surface convex of the fabric. Figure 4 shows the
reflected images of twill weave as a function of the angle of the
incidence, after passing through the image processing system.
Figure 4(a) shows the surface pattern obtained for 10° in-
cidence. The surface contour pattern is not obtained clearly in
this geometry. Figure 4(b) shows the surface pattern for 70°
incidence. The collimated light was incident perpendicular to
the warp yarns. Only warp faces were observed as a bright spots.
The image was not affected by the light intensity diffused from
the filling yarns. The above mentioned results indicate that the
surface contour pattern of the woven fabric is emphasized if the
incident light is inclined or if the sample is tilted. This is mainly
explained as the law of the reflection at the surface of the
cylindrical yarns. The two-dimensional shapes of the light spots
are not always the regular periodic arrangement of point in
space. The light spots are influenced by the yarn geometry such
as the yarn fuzziness, yarn twist and yarn crimp. Figure 4(c)
shows the surface pattern observed at 70° incidence perpen-
dicular to the filling yarns. The surface contour pattern becomes
blurred in comparison with that in Fig. 4(b). The reduction of
the difference between the light and dark level is due to the
decrease in the crimp amplitudes of the filling yarn.
The similar results were obtained in the case of the derivative
weave which contains a small number of the harness. The
surface pattern of the twill derivative is shown in Fig. 5. To a
considerably high extent, either warp faces or filling faces can

Fig. 4 Reflected image patterns as a function of incident angle. (a)10°


incidence. (b) 70° incidence perpendicular to warp yarns. (c) 70°
incidence pararell to warp yarns. All patterns were observed at
right angle with respect to the fabric surface.

Fig. 5 Reflected image of a derivative weave for 80 incidence.

2 Journal of the Textile Machinery Society of Japan


Fig. 6 Regular weave types displayed on TV screen; (a) plain , (b) twill and (c) satin.

Fig. 7 Power spectrum patterns of regular weaves calculated by Fast Fourier Transform method . (a) plain. (b) twill. (c) satin.

be observed as a light spot. The computational analysis for the mercial woven fabrics.
derivative weaves, in practice, needs a lot of pixels, because of Figure 6 shows the reflected images of the three types of the
the high density cloth counts in the structural unit. complete weaves: (a) plain, (b) twill and (c) satin. The light spots
The surface contour pattern of an ideal woven fabric is represent the surface patterns of the warp faces. Each pixel was
composed of identical structural units arranged on a lattice pre-set with a gray level of 64 as a threshold value. Figure 7
defined by two fundamental translational operations. Such a shows the Fourier transform patterns of the reflected images.
structural unit arrangement looks the same in every respect. The The rectangle surrounded by dashed lines indicate the unit cell
lattice defined by these translational operations is a regular in Fourier space. The side length of the unit cell corresponds to
eriodic arrangement of points in space. For convenience, the the pick spacing between the neighboring two yarns. The power
structural unit of the regular weave can be obtained by finding spectrum arrangement (*) is directly compared with the struc-
out the neighboring bright spots on the rectangular coordinates tural unit of weave type.
of the TV screen. It is rather difficult to recognize the repetition Figure 7 (a) shows the Fast Fourier Transform of plain weave.
pattern of the structural unit in the case of the commercial woven Plane weave is characterized by a central spectrum in the unit
fabric, since the woven structures are not always homogeneous. cell. The Fast Fourier Transform pattern is apparently
The periodic rearrangement of the bright spots and the reform- equivalent to the texture pattern of the structural unit. Figure
ing of the blurred shapes are previously required to make the 7(b) shows the Fast Fourier Transform pattern of twill weave.
computational analysis~lol There are two main power spectra indicated by open circles
We will discuss Fourier-pattern sampling as a basis for inside the unit cell. Any other power spectra scattered around
automatic pattern recognition in the following. The Fast Fourier open circles are attributed to the irregular periodic arrangement
Transform is a computational tool which facilitates signal and of the yarn faces. Figure 7(c) shows the Fast Fourier Transform
analysis such as power spectrum analysis and filter simulation pattern of satin weave. As a sample, we chose the satin weave
by means of digital computer. There are two fundamental rela- of 5-harness and 2-counter. Since the dimensions in Fourier
tions which we frequently use in the Fourier analysis. One is the space are reciprocal elements of the spatial dimensions, it is
similarity theorem and the other is the shift theorem. inappropriate to the direct comparison of the texture patterns
The similarity theorem indicates that a scale change in one between the unit cell and structual unit. Comparison was made
domain causes an inverse change in the other. That is, the between the structural unit and the unit cell rotated by through
structural unit made of the crossing combination of the warp and 90 degree. The different image patterns were obtained by choos-
filling yarns is observed in the unit cell as- sociated with the ing the threshold level of the power spectrum. The scattered
neighboring two warp yams and filling yams in Fourier space. spectra comes to disappear if the high threshold value is applied.
It is possible to reduce the detection area scanned on TV screen. The threshold value depends on the irregularity of the warp face
The shift theorem indicates two-dimensional translational arrangement. The relationships between the local irregularity
operation such as the structural unit arrangement looks the same and Fourier analysis have been discussed in the reference)21.
in every respect. The amount of the fundamental translation is The numerical results of Fast Fourier Transform can be
equal to the number of the harness in the structural unit. Figures summarized by pointing out the following three relationships:
6 and 7 show the results of the Fourier transform of the com- (1) The structural unit of weave type is observed in the unit

Vol. 35, No. 2 (1989) 3


cell associated with the neighboring two warp yarns and two [4] Chen Guo Ping, M. Kinoshita, R. Akiyama and S.
filling yarns in Fourier space. Uchiyama: Sen-i Gakkaishi 42, T-528 (1986).
(2) The number of the harness, n, is determined by counting [5] F. Matsuhashi and A. Shinohara: Sen-i Gakkaishi, 21, T-5
the number of the light spots, t, inside the unit cell. The relation- (1965).
ship is given as the form of n = t+l. [6] M. Tanaka: Sen-i Gakkaishi, 7,196 (1951).
(3) The texture pattern of the structural unit is obtained by [7] Y. Goh, F. Matsuhashi and A. Shinohara: Sen-i Gakkaishi,
the Fourier transformation pattern titled clockwise by 90 degree. 17, 989 (1961).
[8] K. Mihira and K. Sumitani: Sen-i Kikai Gakkaishi, 33, T-68
References (1980).
[1] R. Akiyama and E. Kuze: Sen-i Gakkaishi (in Japanese), [9] R. Akiyama, T. Iguro, S. Uchiyama, Y. Magoshi and T.
33, T-439 (1977). Sakai: Sen-i Gakkaishi 42, T-574 (1986).
[2] R. Akiyama, T. Iguro, S. Uchiyama and H. Noshi: Sen-i [10] K. Ohta, K. Sakaue and H. Tamura: Sen-i Kikai Gakkaishi.
Gakkaishi 42, T-521(1986). 36,10 P 459 (1983).
[3] R. Akiyama and E. Kuze: Sen-i Gakkaishi (in Japanese),
35, T-202 (1979).

It is worth your while to have an


accurate, first-hand knowledge of the
constant technological progress in
Japan's textile acid textile machinery
industries.
THE JOURNAL OF THE TEXTILE MA-
CIILNERY SOCIETY OF JAPAN (English edi-
tion) is a medium, in the international language,
of getting that knowledge.

'the Textile Ma chinery Society of Japan is essentially an academic organ devoted to contribution to
the advancement of knowledge and techniques in the textile and textile machinery fields.
Its membership comprises universities , research institutes, textile manufacturing companies, textile
machinery manufacturing companies, bussiness firms with textile and/or textile machinery departments ,
or individuals of expert knowledge and experience connected with these organizations .
The Society takes a modest pride in the contributions it has made to the textile and textile machinery
industries.
As part of its work, the Society publishes a monthly journal in Japanese , setting forth the results of
researches, studies and experiments by its members .
The English edition contains translations of those articles and treatises in the Japanese edition which
the editors believe to be of special international interest .

THE TEXTILE MACHINERY SOCIETY OF JAPAN

Osaka Science &


Technology Center Bldg .,
(Utsubo Park),
8-4, Utsubo-honmachi 1-chome
, Nishi-k,
Osaka, Japan

4 Journal of the Textile Machinery Society of Japan