Technical Journal, University of Engineering and Technology (UET) Taxila, Pakistan Vol. 25 No.

2-2020
ISSN:1813-1786 (Print) 2313-7770 (Online)

A Method for the Classification of Fabric Hand

Based on it Drape
A. Hussain1, G.A. Baig2, M.I. Sarwar3, D. Iqbal4, T. Sultan5, A. Latif6
1,2,5,6
Bahuddine Zakariya University, Multan, Pakistan,
3,4
Central Cotton Research Institute, Multan, Pakistan.
6
[email protected]

Abstract- In this paper fabric hand was objectively and correlated with fabric flexural rigidity by the
evaluated through the exploration of three-dimensional scientists elsewhere [5]. However, DC’s inadequacy to
drape. Four Windows Kinect sensors were arranged to completely account for drape-shape requires exploring
capture the point cloud data of the draped fabric. To more drape-related parameters that may correlate better
extract the actual drape features, a new algorithm was with the complex phenomenon of fabric hand.
designed and implemented in Matlab technical The fabric drape has long been realized a 3D
computing environment. Thirty seven woven fabrics phenomenon. Therefore, in recent years emerging
were clustered in accordance with subjective hand technologies have been taken advantage in determining
preferences of softness and stiffness collectively as 3D drape of fabrics [6]. For instance, 3D scanned images
ground truth. The performance of the objectively have been used to measure the drape parameters for
clustered data was evaluated against the ground truth. fabric [7]. However, due to the older measuring principle
The results revealed that the objective classification of of drape, non-significant results were obtained. A new
the fabrics using 3D drape data correlated to the fabric measuring principle in which force of gravity acts
hand preferences with 86 % matching accuracy when perpendicular to fabric plane was proposed along with
applied on same fabrics. The evaluation of hand from new characterizing parameters [8]. These parameters
fabric drape in a non-touch manner will be a better were claimed to be more conforming with actual
reference for online shopping and computer simulation. aesthetics of fabric drape.
On the other hand, physical instruments like
Keywords- Fabric hand, Fabric classification, 3D Drape PhabrOmeter along with fabric linear density was
model, Principal Component Analysis, Drape. proposed as an alternate test to fabric drape measurement
[9]. Physical involvement of instruments for the
I. INTRODUCTION evaluation of fabric hand is the hindrance to accept these
methods for online customers. Therefore, there is a need
In recent years, online business is booming due to to delve deep into vision perspective (3D drape) for the
its convenience and economy. Contrarily, online sale and objective classification of fabric samples based on hand
purchase of fabrics are restricted due to virtual attributes.
environment of the e-business. The real-time fabric hand In the research work there has been suggested a new
sensation is considered to be more satisfactory as approach derived from 3D drape for the classification of
compared to virtual assessment. For online-shopping, fabrics based on hand. Three dimensional drape-images
evaluation of fabric hand is still a looming problem as the were captured by Kinect depth sensors. The numerical
fabrics and garments cannot be touched. The non- data obtained from depth sensors were processed in
availability of any acceptable and viable method to software to reconstruct the 3D surface. Drape models
classify fabrics based on its hand requires a robust and were analyzed with newly developed algorithms to
acceptable solution. extract drape indicators. The objectively acquired drape
Drape is an important and complex phenomenon of data were classified using Hierarchical clustering
fabric appearance. Drape Coefficient (DC) i.e. the two technique and validated with the ground truth.
dimensional area under the draped fabric, peaks, their
intensity and shapes are considered to be the most II. MATERIALS AND METHODS
important parameters that define the fabric drape [1, 2].
Over the years, efforts have been made to classify fabrics Thirty seven commercially available woven
using drape parameters obtained by Cusick drape tester fabrics were collected from the market. The circular
[3]. Though from 1950 to 2013, almost thirty six drape fabric samples having diameters of 240 mm were hung
indicators have been described in the literature, DC still freely in umbrella shape under gravitational force on the
continues to be the most widely used and acceptable disk (diameter 120 mm) placed on a tripod. The 3D
drape parameter [4]. A new drape coefficient, dependent fabric-drape positional numerical data was generated by
upon draped fabric shape and drape angle was presented special arrangement of Windows Kinect sensors. The 3D

121
 Technical Journal, University of Engineering and Technology (UET) Taxila, Pakistan Vol. 25 No. 2-2020
ISSN:1813-1786 (Print) 2313-7770 (Online)

images of fabrics were processed in Geomagic software well as the methods of handling the samples for the
(3D Systems, USA) to get refined polygonal mesh. The required subjective hand preferences [14-16]. On the
main steps of the current study are shown in Fig. 1. other hand, to assess

A. Scanning of 3D drape
Windows Kinect sensors were arranged to capture point
cloud data from the actual shape of 3D drape. The
computer-controlled sensors use infrared-waves to
capture the image-depth. Kinect sensors require scanning
the image at different locations to capture 3D view.
Therefore, four Kinect sensors placed at corners of the
square table were used to scan 3D drape of the fabric.

B. Registration
Kinect sensors use RGB and infrared sensors to capture
3D image. Coordinates of these sensors need to be
unified for better image quality. Therefore, registration
using Smisek et al.’s algorithm was performed to align
both the images (RGB & IR) [10].

C. De-noising and 3D Surface Reconstruction

The captured 3D drape cloud data were processed in
different software packages. First, the data were
imported into the Geomagic software (3D Systems,
USA) for noise removal and polygonal mesh refinement
as shown in the Fig. 2 a & b.

D. Slicing and Feature Extraction

An algorithm was designed and implemented in the
Matlab software (MathWorks, USA) for the processing Fig. 2: Slicing of three dimensional drape model, (A) 3D
of 3D drape images. Slices were drawn automatically on drape model with slice curves, (B) slice curves and
the surface of 3D drape model as shown in Fig. 2a. Drape boundary curves.
indicators were then derived for the further analysis of samples objectively, four drape variables i.e., drape
the data. To understand the multidimensionality of the coefficient, minimum drape, average trough and drape
drape data, principal component analysis (PCA) was height were assessed through Hierarchical clustering.
performed to increase the perception of data with
minimum loss of information [11, 12].
F. Cluster Analysis
Polythetic Aglomerative Hierarchical Clustering
E. Subjective and Objective Assessment
(PAHC) technique was chosen to classify the 3D drape
Samples measuring 300×300 mm were taken for the
data objectively. In principle, the algorithm starts from
subjective assessment of fabric subjective softness and
singleton-clusters and merge those clusters with minimal
stiffness [13]. A training session was organized to help
distances until all objects are included in one cluster. The
the assessors become familiar with the descriptors as

Scanning Denoising Slicing Validation

Subjective and Cluster of
and and
of Registration Objective
3D surfacee Feature Assessment analysis 3D Drape
3D drape reconstruction Extraction data

Fig. 1: Main Stages involved in the determination of fabric objective hand

122
 Technical Journal, University of Engineering and Technology (UET) Taxila, Pakistan Vol. 25 No. 2-2020
ISSN:1813-1786 (Print) 2313-7770 (Online)

average linkage (L) is defined below mathematically in terms of softness and stiffness. Two principle
(1), components explaining 95.54% data variability were
selected.
1
L = |𝐴||𝐵| ∑𝑎∈𝐴 ∑𝑏∈𝐵 𝑑(𝑎, 𝑏) (1)
Samples requiremnet and data sphericity (homogeneity
of variances) results were validated with Kaiser-Meyer-
where d selected as metric (Euclidian distance); 𝑑(𝑎, 𝑏) Olkin Sampling Adequacy test and Bartlett's Test of
is the distance between two elements of clusters A and Sphericity. The value of Kaiser-Mayer-Olkin sampling
B. adequacy test was observed 0.83 (more than 0.5) proved
that the selected number of samples (37) are statistically
The clustering procedure is completed in an n-
large enough for further processing of the data with PCA.
dimensional space (n is the number of variables); this
Results of the subjective hand preferences of fabric
means during clustering the difference of every
softness and stiffness classified the samples into three
parameter between samples is considered, so the results
clusters, representing the classes of fabrics viz., class
are much more realistic and definite, this being a unique
1(soft), class 3 (moderately stiff) and class 2 (stiff) as
advantage of clustering compared to other statistical
shown in column 2 of the Table II.
approaches [17].
To match fabric preferences objectively with subjective
Requirements of data and sphericity (homogeneity of
hand preferences, drape data were processed through
variances) for further processing of data through
PCA and clustered. Interestingly, the objective data were
principal component analysis was checked with Kaiser-
also classified into three classes as was the case with
Mayer-Olkin test and Bartlett's test respectively, as
fabric hand preferences as shown in Table II.
shown in Table I.
TABLE II. CLUSTER VALIDATION FOR FABRIC HAND AND
DRAPE
TABLE I. KMO AND BARTLETT'S TEST OF 3D DRAPE DATA
Cluster Numbers
KMO and Bartlett's Test Sample No.
Fabric Hand Fabric Drape
1 1 1
Kaiser-Meyer-Olkin Sampling
0.83 2 1 1
Adequacy
3 2 2
4 2 3
Bartlett's Test of Sphericity Sig. 0.000 5 3 3
6 3 3
7 3 3
G. Validation of 3D-Drape Data
8 3 3
For the validation of results, an external criterion was
9 3 3
adopted which measured the performance of the PAHC
technique in terms of matching accuracy. Let xi denotes 10 3 3
the number of objects correctly matched to the ith cluster 11 2 2
Ci; f is matching accuracy, mathematically is described 12 1 1
as follows, 13 1 1
1
14 3 3
𝑓 = ∑𝑘𝑖=1 𝑥𝑖 (2)
𝑛 15 3 2
16 2 2
where k is the number of clusters and n is the total
17 2 2
number of samples assessed.
18 2 2
19 2 2
III. RESULTS AND DISCUSSIONS 20 2 2
PCA was applied on drape parameters to classify 21 2 2
fabrics objectively with the fabric hand preferences in 22 3 2

123
 Technical Journal, University of Engineering and Technology (UET) Taxila, Pakistan Vol. 25 No. 2-2020
ISSN:1813-1786 (Print) 2313-7770 (Online)

23 2 2 Fig. 3. Pairwise matching accuracy of clustered values

of fabric hand and 3D drape.
24 2 3
25 2 2 IV. CONCLUSIONS
26 3 2
This paper presented efficaciously an objective
27 2 2
method for the classification of fabric-hand derived from
28 2 2 3D numerical datasets of fabric drape. The fabrics were
29 2 2 scanned to get 3D drape models using a method
30 2 2 developed in our earlier research with the low-cost
31 3 3 devices. Objective classification of fabrics based on 3D
drape significantly matched with subjective hand
32 2 2
assessments. Fabrics were represented by three classes.
33 1 1
In future, more 3D drape indicators need to be explored
34 1 1 to enhance the matching accuracy of fabric drape with its
35 1 1 hand.
36 1 1
37 1 1
Matching REFERENCES
32 / 37 = 86%
Accuracy
[1]. G. Cusick, 1965, The Dependence of Fabric
Drape on Bending and Shear Stiffness, J. Text.
To validate the clustered data (subjective hand
Ins., 56(11), pp. T596-T606.
preferences and 3D drape) matching accuracy of the
[2]. E. Buyukaslan, S. Jevsnik, and F. Kalaoglu,
objective (PCA) classification with subjective hand
2018, Comparative analysis of drape
preferences of fabric softness and stiffness combined was
characteristics of actually and virtually draped
examined. The overall matching accuracy using (2) was
fabrics. Int. J. Cloth. Sci. Tech., 30(3), pp. 286-
observed f = 86% i.e. 32 out of 37 samples matched
301.
correctly with the objective clustering of 3D drape in
[3]. L. Sang-Song, 2004, Using drape form to
respective clusters. The entire sample was divided into
establish discriminant models of fabric
three hierarchical clusters objectively, as was the case of
characteristics, Indian J. Fibre Text. Res., 29(2),
ground truth, shown in Fig. 3 and Table II. Cluster 1
pp. 143-148.
represented the soft, cluster 2 represented the stiff and
[4]. E. Carrera-Gallissà, X. Capdevila, and J.
cluster 3 represented the moderately stiff fabrics.
Valldeperas, 2017, Evaluating drape shape in
Overall, five out of thirty seven specimens were miss
woven fabrics, J. Text. Ins., 108(3), pp. 325-
classified in different clusters as shown in Fig. 3 and
336.
Table II. It means fabric hand from 3D drape was
[5]. E. Messiry, M. and S. El-Tarfawy, 2020,
successfully classified based on the customer subjective
Investigation of fabric drape-flexural rigidity
preference.
relation: modified fabric drape coefficient. J.
Text. I., 111(3), pp.416-423.
[6]. Y, Zhicai, Z. Yueqi, R. H. Gong and X.
Haoyang, 2019, New indicators on fabric drape
evaluation based on three-dimensional
model, Text. Res. J., p.0040517519888669.
[7]. V. Glombikova, and Z. Kus. 2014, Drape
Evaluation by the 3D Drape Scanner, Tekstil ve
Konfeksiyon, 24(3), pp. 279-285.
[8]. Z. Mei, W. Shen, Y. Wang, J. Yang, T. Zhou,,
& H. Zhou. 2015, Unidirectional Fabric Drape
Testing Method, PloS one, 10(11), e0143648.
[9]. N., Pan, C. Lin, and J. Xu, 2019, A new method
for measuring fabric drape with a novel

124
 Technical Journal, University of Engineering and Technology (UET) Taxila, Pakistan Vol. 25 No. 2-2020
ISSN:1813-1786 (Print) 2313-7770 (Online)

parameter for classifying fabrics. Fibers, 7(8), [14]. H. S. Ryu, and E. K. Roh. 2010, Preference and
p.70. subjective evaluation of washed fabric hand
[10]. J. Smisek, M. Jancosek, & T. Pajdla. 2013, 3D using conjoint analysis, Text. Res. J., 80(20), pp.
with Kinect. In: Fossati A, Gall J, Grabner H, 2167-2175.
et al. (eds) Consumer depth cameras for [15]. E. Bertaux, S. Derler, R. M. Rossi, X. Zeng, L.
computer vision. London: Springer, pp. 3-25. Koehl, & V. Ventenat, 2010, Textile,
[11]. D. F. Morrison, 1990, Multivariate statistical physiological, and sensorial parameters in sock
methods, New York, NY. Mc, comfort, Text. Res. J., 80(17), pp. 1803-1810.
[12]. H. Kanai, M. Morishima, K. Nasu, T. [16]. D. Grinevičiūtė, and M. Gutauskas, 2004, The
Nishimatsu, K. Shibata & T. Matsuoka. 2011, comparison of methods for the evaluation of
Identification of principal factors of fabric woven fabric hand. Materials Science
aesthetics by the evaluation from experts on (Medžiagotyra), 10(1), pp. 97-100.
textiles and from untrained consumers, Text. [17]. N. Pan, K.C.Y., S. J. Zhao, and S. R. Yang,
Res. J., 81(12), pp. 1216-1225. 1988, A New Approach to the Objective
[13]. P. Zhang, X. Liu, L. Wang, & X. Wang, 2006, Evaluation of Fabric Handle from Mechanical
An experimental study on fabric softness Properties Part III Fuzzy Cluster Analysis for
evaluation, Int. J. Cloth. Sci. Tech., 18(2), pp. Fabric Handle Sorting, Text. Res. J., 58(10), pp.
83-95. 565-571.

125