Cashmere Lab Description

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Measurement 185 (2021) 109991

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Calibrated color measurement of cashmere using a novel computer


vision system
Chong Heng a, b, Hua Shen a, b, Fumei Wang a, b, *, Yufeng Chen c, d
a
College of Textile, Donghua University, Shanghai 201620, China
b
Key Laboratory of Textile Science & Technology Ministry of Education, Shanghai 201620, China
c
College of Textile and Clothing, Xinjiang University, Xinjiang 830046, China
d
China Fibre Inspection Bureau, China

A R T I C L E I N F O A B S T R A C T

Keywords: Color of cashmere fiber is a key component of textile quality inspection. However, the inspection of cashmere
Cashmere color color was determined by human vision system, which was less accuracy and time-consuming. Computer vision
Computer vision system (CVS) is considered as a promising technique to objectively and precisely test color. In the present work, a
Image analysis
novel color measurement system for cashmere was proposed. Totally 29 cashmere samples with different color
Calibration model
Classification
were adopted as standard samples to calibrate color conversion model. The correlation coefficient of L, a, b
values between the two systems was separately calculated high to 0.99, 0.96 and 0.93 for the whole samples. The
proposed method was further validated by other 15 samples, indicating the accuracy of the novel CVS. Besides,
due to the high accuracy and strong representativeness of the new method, the categories of cashmere, which
were normally tested by subjective visual assessment, could be determined by the present results.

1. Introduction procedures undertaken for the evaluation of quality of cashmere are


done organoleptically. Cashmere was divided into three categories
Cashmere fiber, known as “soft gold”, constitute a crucial element of including white, grey and brown according to its color, which is esti­
the human clothing [1,2]. Color is the initial feature of cashmere judged mated by visual inspection [6]. Such visual inspection is subjective, less
by consumers [3,4], and therefore is a key grading factor for most textile efficient and time-consuming, since it needs rich experience to guar­
products. “Whiteness” is usually used to evaluate the color of textile antee the precise of color categories determination [7] and calling for
materials, which is a one-dimensional parameter of the color of material accurate instrumental measurement. However, any discussion in the
surface. In the process of fiber dyeing, color difference exists in fibers literature about cashmere fibers color measurement using instrumental
with same whiteness which are dyed with the same process because method has been lacking.
whiteness cannot accurately and comprehensively characterize the color Several available techniques have been proposed to resolve the
of fibers. Therefore, it is reasonable to evaluate fiber color in three- above drawbacks, including traditional instruments like spectropho­
dimensional color space. Three alternatives (Lab, RGB and XYZ space) tometer [8–10] which has been utilized in other fields than cashmere.
can be selected to define and quantify the color of textile fiber. Lab space However, for the color measurement of cashmere, the area under
is the most used one because of its uniform in describing color. There­ observation is not big enough (only a few cm2) to represent the color of
fore, it is selected to represent the color of cashmere fiber. In Lab space, whole cashmere [11] due to its high inner color variation. Hence, to
it has three channels L, a and b. Parameter L describes the lightness of further improve the precision of measurement, repetitive observations
the sample, locating the black to white in the grey axis. a and b sepa­ are necessary but a time-consuming process. Besides, spectrophotometer
rately describe chromatic color, from green to red and from blue to only provides average color values of small areas, which lacks the ability
yellow. of capturing color information of each pixel, and therefore, colorimetric
Currently there is no quantitative and standardized instrumental characteristics is unrepresentative and uncomprehensive. Cui et al. [12]
method for measuring and analyzing the color of cashmere. In the reported that cottons with the same overall color values could exhibited
Chinese national standard GB/T 18267-2013 [5] Document, any different intra-sample color variation, indicating intra-sample variation

* Corresponding author at: College of Textile, Donghua University, Shanghai 201620, China.
E-mail address: [email protected] (F. Wang).

https://doi.org/10.1016/j.measurement.2021.109991
Received 7 May 2021; Received in revised form 1 July 2021; Accepted 2 August 2021
Available online 20 August 2021
0263-2241/© 2021 Elsevier Ltd. All rights reserved.
C. Heng et al. Measurement 185 (2021) 109991

was important in grading cotton. Additionally, high cost and repair fee Besides, other 15 samples (classified into 4 white, 5 grey and 6 brown)
of spectrophotometer also limit its application in some textile mills. were used to validate the proposed the calibrated CVS.
Instead, computer vision system (CVS) [13,14], which is simple, fast,
unexpensive and strong representative, has become a common and 2.2. Methods
affordable means in many measurement fields such as food (meat [15],
potato chips [16], bread [17]), agriculture (cotton [12,18], rapeseeds The precise and cheap CVS, which can achieve image capturing
[14], vegetables [19,20], tomato [21], rice[22]) and medical (teeth (Fig. 1a) and image analysis (Fig. 1b), was designed to measure the color
[23], cancer diagnosis [24], Chinese medicine [25], tongue [26]). In of cashmere. Fig. 1(c) showed the measurement process of standard
detail, this method captures the image containing color information colorimetric characteristics (RGB, XYZ and Lab), which was utilized to
with millions of pixels, which enables further analysis for color varia­ calibrate and validate the present method.
tion. However, the measurement working based on image analysis is still
a great challenge resulted from changes in cameras, illuminations and 2.2.1. Computer vision system
detection regions which cause color variations for the same scene. In
other words, it is a key point to calibrate the algorithm to evaluate (1) Image capturing
fibrous cashmere color in textile quality assessment [27]. Besides, for
the measurement of cashmere material, a series of parameters need to be Color testing results of fibrous material will be affected by factors
manipulated to standardize the measurement [12]. Therefore, precise including temperature/humidity, thickness and applied pressure on
and easy-to-use techniques based on CVS should be developed. cashmere biscuits, so standardization of its experimental conditions was
In this study, we proposed a calibration model to calculate color investigated. The detailed influence exerted from the above factors was
values by image analysis, to eliminate error caused by devices. The carefully discussed in section 3.1. Therefore, before image capturing, 20
calibrated CVS could be adopted to provides good-representation and g sample was evenly put into the box in standard constant temperature
accuracy for color measurement and quantitative indexes for the and humidity room (20 ± 2 ℃, 65 ± 4 % RH), and then 4 kg pressure
assessment of color categories. Moreover, experimental conditions for was applied on the biscuit according to the present study.
cashmere color measurement were determined to standardize its color As shown in Fig. 1a, the device for image capturing includes light
testing. Therefore, this study was implemented with following goals: (i) emitting diodes with 6000 K color temperature (LEDs, 3 W each, 30 cm
explore the standard experimental condition for cashmere color mea­ length), charge-coupled device (CCD) camera (Sony, ICX282AQ, 20.48-
surement to make the results accurate and consistent. In this part, three megapixel CCD sensor), mirrors and sample box (12 × 12 × 6.5 cm). LED
factors including thickness, applied pressure of samples and tempera­ lamps were arranged with a distance of 4 cm away from the glass and an
ture/humidity were studied based on color values; (ii) for the accuracy angle of 45◦ to provide a uniform illumination [29,30]. A digital CCD
of color measurement, CVS was calibrated based on the relationship of camera, the cheapest image capturing component [31], with high res­
color values obtained by spectrophotometer; (iii) to objectively and olution and low noise was used to capture bitmap-formatted images
quantitatively evaluate the color categories of cashmere, the color with 1000 dpi and has been set up with automatic white balance,
values distribution of cashmere in Lab space was analyzed. aperture F/10.0, 1/8 sec exposure, flash off. During the measurement,
the testing sample was evenly and randomly placed in a box, above
2. Materials and methods which a slider was applied to generate a constant 4 kg pressure on the
sample. Since loosely packed structure will increase the light trans­
2.1. Samples mission into the biscuits, and consequently reduce the light reflection
detected by the sensors, which would affect the final testing results. The
Totally 29 cashmeres varying colors (purchased from a commercial data obtained from the central region image (10 × 10 cm) was selected
distributor in Erdos China) were selected as standard samples that they to calculate color values (RGB, XYZ and Lab). The area under observa­
were used to calibrated the CVS system, and their color values were tion in the present work which was much higher (~14 ~ 300-fold) than
tested by Datacolor [28]. Color categories were tested by trained that in common spectrophotometer. Additionally, the non-detection
workers (Erdos company in China), and 9 samples, 8 samples and 12 area was covered with a black shading mask to minimize the influence
samples were separately classified into white, grey and brown cashmere. of internal and external light.

Fig. 1. Methodology for the color measurement. a ~ b Illustration of computer vision system. c Standard colorimetric values (RGB, XYZ, Lab).

2
C. Heng et al. Measurement 185 (2021) 109991

(2) Image analysis 10. Validation of the proposed calibration model for cashmere color
measurement using new samples.
After RGB digital image captured by CCD camera, a logical proced­
ure for image analysis was carried out on a Graphical User Interface 3. Results and discussion
(GUI) created in MATLAB. R, G and B values of the selected saving re­
gions were conducted by MATLAB 2012a software. Color values of 3.1. Determination of experimental conditions
different spaces (Lab, XYZ and RGB) were obtained according to the
traditional color conversion [17,32]. The dependence of color values of the selected samples on factors
including thickness, applied pressure and testing environment was
2.2.2. Standard colorimetric characteristics determined according to the Methodology in Fig. 3. For the color mea­
The standard color characteristics for the calibration and validation surement of fibrous material, proper experimental conditions were a key
of CVS, RGB, XYZ and Lab values, were measured using a spectropho­ step to ensure the extraction of “true” cashmere color. Therefore, it is
tometer (Datacolor-850, D65 lamp; 10◦ observer; 3 cm aperture) in a necessary to investigate how the true color of cashmere fiber can be
standard constant temperature and humidity room (20 ± 2 ℃, 65 ± 4 % tested instead of comparing with standard color values, and the values in
RH). During the testing process, 6 g sample was packed tightly into the Fig. 4 and Fig. 5 were not calibrated values before the calibration of CVS
same compression density and thickness as performed using the exper­ system. Since cashmere fiber is translucent material, and its color ob­
imental device, ensuring that the light from lamp did not penetrate the tained from image analysis is usually influenced by the above three
cashmere fiber biscuits. Values were recorded when the error rate was factors. Under different temperature/humidity levels, the microstruc­
less than 3 %. For a given sample, the testing numbers were separately 3 ture of cashmere changes, resulting in the difference of the interaction of
~ 5 and 5 ~ 20 times for white cashmere, grey and brown cashmere. light which was expressed in the captured image finally. The moisture
regain rate of cashmere is 17.0 % and it gradually enhanced with the
2.2.3. Color calibration process increasing of air humidity, which further resulted the decrease of luster
The methodology illustrated in Fig. 2 was utilized in this study to and the darkening of color. Except for testing environment, applied
develop the calibration model for color conversion. The calibration pressure on cashmere biscuit was also a key factor for color measure­
process was performed as follows: ment. Air mixed in the fibrous biscuit needs to be squeezed out by
pressure to test the actual color, since air leaded to the diffusion of light.
1. 29 cashmeres with different colors were selected as standard Besides, the thickness of biscuit is of great importance for the mea­
samples, aiming to get a wide range to propose a maximum surement, and proper thickness ensures the capturing of real color
adaptable model. without background. While the influence from the above factors on
2. Selection of 3 samples including a white, a grey and a brown
cashmere among 29 cashmeres as the test samples that they were
used to explore the proper experimental conditions. Recording of
images using experimental device under different thickness,
pressure, temperature and humidity conditions.
3. Extraction of RGB values of the above selected 3 samples under
different experimental conditions using MATLAB.
4. Determination of parameters for proper experimental conditions
of the CVS system based on the result of previous step.
5. Images capture of 29 samples under the determined standard
experimental conditions.
6. Extraction of RGB values of 29 samples using MATLAB.
7. Calculation of XYZ, Lab values through the traditional color space
conversion using MATLAB.
8. Measurement of standard color values (RGB, XYZ, Lab) using
spectrophotometer (Datacolor).
9. Development of the calibration color conversion using the poly­ Fig. 3. Methodology for the determination of experimental conditions for
nomial model. sample preparation.

Fig. 2. Steps used to develop calibration model for color conversion.

3
C. Heng et al. Measurement 185 (2021) 109991

b c
10 g R (R) 20 g R (R)
R (S)
R (S)
G (R)
G (R) 200
200 G (S)
G (S)
B (R)
B (R)
RGB values

B (S)
RGB values

B (S)

180 180

160 160
1 4 7 10 1 4 7 10
Pressure (kg) Pressure (kg)
d
30 g R (R)
R (S)
G (R)
200 G (S)
B (R)
RGB values

B (S)

180

160
1 4 7 10
Pressure (kg)
Fig. 4. Images and RGB values of white cashmere under different preparation. a Images of white cashmere. b ~ d RGB values of captured images: 10 g for b, 20 g for
c, 30 g for d.

testing results fails to be considered by researchers [12]. Therefore, the • Weight of samples: 10, 20 and 30 (g)
following three factors with different levels were studied to clarify their • Applied pressure: 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10 (kg)
influence on cashmere color measurement.
The effect of the above three factors on color was studied subjectively
• Testing environment through RGB images captured at different prepared white sample. As
Standard condition (S): constant temperature/humidity (20 ± 2 shown in Fig. 4a, uniform illumination, being clearly demonstrated on
℃/65 ± 4 % RH) images, ensured extraction of color information from cashmere. By
Room condition (R): non-constant temperature/humidity (25 ± 8 observing the captured images, color was obviously influenced by
℃/80 ± 10 % RH) testing environment than pressure and thickness. Fig. 4b~d, the

4
C. Heng et al. Measurement 185 (2021) 109991

b c d
10 g R (R) 20 g R (R) 30 g R (R)
R (S) 200 R (S)
200 R (S) 200 G (R) G (R)
G (R)
G (S) G (S) G (S)
B (R) B (R) B (R)

RGB values
RGB values
RGB values

B (S) B (S) B (S)


180 180 180

160 160 160

140 140 140

1 4 7 10 1 4 7 10 1 4 7 10
Pressure (kg) Pressure (kg) Pressure (kg)

f g h
180 180 180
10 g R (R)
R (S)
20 g R (R)
R (S) 30 g R (R)
R (S)
G (R) G (R)
G (R)
G (S) G (S)
G (S)
160 B (R) 160 B (R)
160 B (R)
RGB values

RGB values

B (S)
RGB values

B (S) B (S)

140 140 140

120 120 120

1 4 7 10 1 4 7 10 1 4 7 10
Pressure (kg) Pressure (kg) Pressure (kg)

Fig. 5. Images and RGB values of grey and brown cashmere under different preparation. A Characterizations of grey cashmere. a Images of grey cashmere. b-d RGB
values of grey cashmere: 10 g for b, 20 g for c, 30 g for d. B Characterizations of brown cashmere. e Images of brown cashmere. f-h RGB values of brown cashmere: 10
g for f, 20 g for g, 30 g for h.

5
C. Heng et al. Measurement 185 (2021) 109991

behavior of R, G, and B, gotten by analyzing the captured images also Table 1


confirmed this tendency. In detail, temperature/humidity had the . The differences of RGB values under the black and white background.
greatest influence on the R index, followed by G and B in turn. Compared Thickness/mm ΔR ΔG ΔB
with the results obtained from standard condition, the RGB values
9 3.370 2.223 0.070
showed monotonic increasing under room condition at the same pres­ 15 0.014 0.001 0.001
sure and thickness. For all plots (Fig. 4b~d), the RGB values showed 20 0.014 0.001 0.000
monotone increasing in response to the raising of pressure from 1 kg to 4
kg and almost remained unchanged over 4 kg due to the structural
change of fiber biscuits from loose to tightness during the increasing of occurred when RGB was converted into Lab space using the direct for­
pressure. In the early stage, the air inside the fiber biscuit was squeezed mulas. As shown in Fig. 6, the direct Lab values obtained by CVS were
out below the threshold leading to the arrangement of the fibers closer. compared with that obtained by Datacolor. For the whole set of cash­
Then the real color of the cashmere fiber was captured when reaching mere samples, different relationships were exhibited for the three in­
the threshold. dexes, being the coefficients of determination (R2) equal to 0.986, 0.518
The similar tendency was also observed from the selected grey and and 0.623 for L, a and b, respectively. L showed the best agreement than
brown samples. Fig. 5a~d and Fig. 5e~h represented the results of grey parameter a and b, which is in accordance with previous work. For
and brown cashmere, respectively. However, unlike white cashmere, B example, Vilbett Briones [33] studied the use of digital technology to
values of grey and brown cashmere were almost independent by tem­ test the color of chocolate, and the agreements of chroma index a and b
perature/humidity, pressure and thickness. Especially for brown cash­ were poor, with the R2 of 0.768 and 0.505, respectively, while that of L
mere, the R, G, and B values remained almost unchanged under two was high to 0.912. In Fig. 6b and c, it’s worth noting that the color of
atmospheric environments, which was consistent with the results of samples whose a and b values appearing not in linear region were darker
subjective images (Fig. 5e). In other words, the lighter color cashmere (brown cashmere), so errors occurred in the darker black samples with
was more sensitive to the above three factors. Therefore, for color the direct conversion model.
measurement of whole cashmere, a proper and constant experimental The accuracy of color measurement based on direct conversion
condition is needed to make the final results stable, and standard con­ model needs to be improved, and calibration is necessary to ensure the
dition (20 ± 2 ℃, 65% ± 4% RH) and 4 kg pressure was recommended application of computer vision in darker cashmere color measurement.
according to the above study. Linear polynomial model (P) with regressors was used to calibrate CVS
Hereinbefore, the color measurement results are also influenced by by observing the relationship of RGB, XYZ values obtained by the two
thickness, as fiber is translucent material and it is possible to see back­ systems, since RGB and XYZ values supplied a CCD color feature which
ground color under insufficient thickness. To determine the threshold of acted as the intermediate model for the conversion of Lab spaces.
thickness above which the color background exert no influence on the CP1 (R, G, B) = [R, G, B, 1]ϑP1 (1)
measurement results, two plates with different color backgrounds (white
and black) were utilized. Color value was the composition of the tested CP2 (X, Y, Z) = [X, Y, Z, 1]ϑP2 (2)
fibers and background environment under insufficient thickness, while 1×4
it showed unobvious change once over the threshold of thickness. A Where C is the output calibrated color values, and ϑP1 , ϑP2 ∈ R are
white cashmere sample was adopted due to its color sensibility to parameters to be determined by polynomial model P1, P2.
thickness, and this experiment was implemented under standard testing The first step for calibration was to investigate the correlation of
environment (20 ± 2 ℃, 65 ± 4 % RH) and 4 kg pressure conditions (as parameters obtained by the two systems, and their relationship was
described in Fig. 4 and Fig. 5). The differences of RGB values, under the utilized to calibrate CVS. Fig. 7a to c demonstrated the plots of RGB
black and white background plates, were compared as shown in Table 1. values obtained by CVS against that of Datacolor on 29 samples, and a
The differences monotonic decreased in response to the thickness values strong agreement among RGB values obtained by the two devices was
of 9, 15 and 20 mm. The difference values of R, G and B were separately clearly observed, being the R2 equal to 0.980, 0.987 and 0.991,
3.370, 2.223, and 0.070 for 9 mm thickness. While for 15 mm and 20 respectively. This showed that the expected linear correction can be
mm thickness, the difference values were separately 0.014, 0.001, and utilized for the calibration of RGB values obtained by CVS. RGB space, as
0.001, 0.014, 0.001 and 0.000. When the thickness of sample was 15 a device-dependent model, performed differently for the same sample
mm, RGB values could be treated as constant because the differences of with different devices. Calibrated Lab values were obtained using P1
RGB values were enough small. That is to say, 15 mm was the threshold model. Eq. (3)–(5) was the calibrated polynomial models (P1) for RGB
of thickness for cashmere fiber color measurement. values, which was developed based on the relationship of RGB obtained
by the two devices, could be utilized to calibrate the results acquired by
CVS into a standard level.
3.2. Calibration and validation of CVS √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
100
× (0.213R* + 0.715G* + 0.072B*)
(3)
3
L = 116 × 255 − 16
Traditional color conversion without calibration is the most y0
straightforward model, which suits for standard system containing ideal
camera, illumination, and observation angle. However, differences exist
among the testing principle, instrument and testing area, so error

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
100 100
× (0.412R* + 0.358G* + 0.180B*) × (0.213R* + 0.715G* + 0.072B*)
(4)
3 3
255 255
a = 500 × ( − )
x0 y0
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
100 100
× (0.412R* + 0.358G* + 0.180B*) × (0.019R* + 0.119G* + 0.950B*)
(5)
3 3
255 255
b = 200 × ( − )
x0 z0

6
C. Heng et al. Measurement 185 (2021) 109991

Fig. 6. Relationship of Lab values obtained by CVS and Datacolor. a L values. b a values. c b values.

a b
250 White cashmere 250 White cashmere
Grey cashmere Grey cashmere
Brown cashmere Brown cashmere
G (Datacolor)
R (Datacolor)

200 200

150 150

100
100 R2=0.980 R2=0.987

50
100 150 200 250 50 100 150 200 250
R (CVS) G (CVS)

c
White cashmere
Grey cashmere
200 Brown cashmere
B (Datacolor)

150

100
R2=0.991

50
50 100 150 200
B (CVS)
Fig. 7. Relationship of RGB values obtained by the two devices.

7
C. Heng et al. Measurement 185 (2021) 109991

a 80
b
80
White cashmere White cashmere
Grey cashmere Grey cashmere
60 Brown cashmere 60 Brown cashmere

X (Datacolor)

Y (Datacolor)
40 40

20 20
R2=0.987 R2=0.987

0 0
0 20 40 60 80 0 20 40 60 80
X (CVS) Y (CVS)
80
c
White cashmere
Grey cashmere
60 Brown cashmere
Z (Datacolor)

40

20
R2=0.987

0
0 20 40 60 80
Z (CVS)
Fig. 8. Relationship of XYZ values obtained by the two devices.

Where tri-stimulus values x0 , y0 and z0 , were separately 95.047, 100


Table 2
and 108.883 determined by the weighted-ordinate method of CIE 1995.
. R2 of Lab values between CVS and Datacolor.
Fig. 8a to c demonstrated the plots of XYZ values acquired by the two
Index
devices. Clear linear agreements of XYZ values were exhibited with R2
higher than 0.98 for all fit functions, which is a basis for the calibration Method L a b
of CVS. Polynomial models (P2) were illustrated in Eqs. (6)–(8). Cali­ Original conversion 0.986 0.612 0.623
brated Lab values were obtained using P2 model, which was the com­ RGB calibration 0.986 0.959 0.916
bination of polynomial models and traditional color conversion. The XYZ calibration 0.987 0.962 0.928
calibration avoided the problem of lower XYZ values obtained by direct
conversion model and the inapplicable region of the conversion from
calibration was much higher than that of direct conversion (from 0.518
XYZ to Lab space.
to 0.962 for parameter a, and from 0.623 to 0.928 for parameter b),
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
131.9 indicating the developed calibrated method was an effective way to
3 165.0 + × (0.213R + 0.715G + 0.072B)
L = 116 × 255
− 16 (6) diminish the systematic errors caused by devices. Therefore, P2 cali­
y0
bration model (Eqs. (6)–(8)) was utilized to automatically test the

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
128.9 131.9
3 250.7 + × (0.412R + 0.358G + 0.180B) 3 165.0 + × (0.213R + 0.715G + 0.072B)
a = 500 × ( 255
− 255
) (7)
x0 y0
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
128.9 131.1
3 250.7 + × (0.412R + 0.358G + 0.180B) 3 255.3 + × (0.019R + 0.119G + 0.950B)
b = 200 × ( 255
− 255
) (8)
x0 z0

cashmere fiber color.


To validate the accuracy of calibration P2 model, 15 new samples
Table 2 showed the correlation of Lab values between the standard were utilized to compare the original 29 samples. Fig. 9 shows that the
color values and calibrated or original values. Compared with the Lab values obtained by Datacolor was closely related to that of cali­
simplest direct conversion model, RGB and XYZ calibration performed brated CVS with R2 high to 0.988,0.966 and 0.961, respectively, indi­
well at the application of CVS in color measurement. Especially XYZ cating the accuracy of the proposed calibration model. Moreover, the
calibration had better effect, and the R2 of Lab values calculated by XYZ practical capturing condition (i.e., illumination and camera) and

8
C. Heng et al. Measurement 185 (2021) 109991

a b
10
90 Samples for calibration Samples for calibration
Samples for validation 8 Samples for validation

a (Datacolor)
L (Datacolor)
75
6

60 4

2
45 R2=0.988 R2=0.962
0
30
30 45 60 75 90 0 2 4 6 8 10
Calibrated L Calibrated a

c
Samples for calibration
18
Samples for validation
b (Datacolor)

15

12
R2=0.956
9

9 12 15 18
Calibrated b
Fig. 9. Estimate of a L, b a and c b for calibrated model for actual vs predicted value.

conditions. Therefore, the calibration equations derived in this study


Table 3 performed well in the color measurement of cashmere fiber.
. Color difference of 15 additional samples between calibrated CVS and
Furthermore, the Lab color differences of additional 15 samples be­
Datacolor.
tween the two devices were calculated by Eqs. (9)–(12), and detailed
Sample No ΔEab Sample No ΔEab Sample No ΔEab results were shown in Table 3. Total color difference ΔEab , which is the
30# 1.69 35# 1.68 40# 1.08 best parameter to represent color difference, was found to range from
31# 0.74 36# 0.46 41# 1.26 0.46 to 1.71 with the average of 1.24, indicating the effectiveness of
32# 1.71 37# 1.23 42# 0.51
calibrated model for cashmere color measurement.
33# 1.17 38# 0.91 43# 1.34
34# 0.79 39# 1.67 44# 0.88 ΔL = |L1 − L2 | (9)

Δa = |a1 − a2 | (10)

90 L Δb = |b1 − b2 | (11)
a
b
Lab Values

60

30 White cashmere Grey cashmere Brown cashmere

0 5 10 15 20 25
Sample No
Fig. 10. Distribution of three categories cashmere in Lab space.

experimental conditions also guaranteed the accuracy of final result. In


detail, lighting setting in our study ensured uniform and consistent
illumination, and the influence of background color and atmospheric
environment was removed due to the standardized experimental
Fig. 11. Variation coefficients of Lab values obtained by calibrated CVS.

9
C. Heng et al. Measurement 185 (2021) 109991

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ University (CUSF-DH-D-2020014).


ΔEab = ΔL2 + Δa2 + Δb2 (12)
Where 1 and 2 represent calibrated CVS and Datacolor, respectively. References
Currently, the cashmere color including white, grey, and brown was
[1] X. LuYanping, Influential Factors of China Cashmere ProductsTrade: An Analysis
determined by trained worker. Based on the color measurement system
Based on CMS Model, Agricultural Economics and Management 4 (2019) 53–61.
proposed in the present paper, the color categories of cashmere can be [2] X. Sun, S. Cheng, H. Yuan, C. Song, L.i. Wang, W. Tian, X. Li, Rapid drying-free
determined objectively, which is more accurate and faster than visual determination of pure cashmere content in scoured cashmere using a novel method
of NIR spectroscopy combined with moisture elimination and spectral
evaluation. Fig. 10 showed that parameter L value of different color
reconstruction strategy, Vib. Spectrosc. 106 (2020) 103006, https://doi.org/
categories of cashmere was different, so their distribution features could 10.1016/j.vibspec.2019.103006.
be used to determine color categories. The L value of white cashmere [3] D. A. Saghi, S. A. Shiri, M. Nikbakhti, Study on Cashmere Quality and Hair
was the largest, followed by grey and brown cashmere. L value of white Percentage of Black Goat of Southern Khorasan for Stable Development of
Agriculture, Pakistan Journal of Nutrition 6 (4) (2007) 397–398.
cashmere ranged from 86.77 to 89.20, and that of grey and brown [4] B.A. McGregor, C. Kerven, S. Toigonbaev, Sources of variation contributing to
cashmere was found range from 74.89 to 82.62 and from 34.85 to 69.50, production and quality attributes of Kyrgyz cashmere in Osh and Naryn provinces:
respectively. However, the a and b value showed no monotonic change Implications for industry development, Small Ruminant Research 84 (1-3) (2009)
89–99, https://doi.org/10.1016/j.smallrumres.2009.06.008.
with the deepening of color. Therefore, the distribution characteristics [5] Cashmere.
of index L can be utilized to quantitatively evaluate the color categories [6] MULS BBS. Scientific journal-agricultural economics. Mongolia: School of
of cashmere. Economics and Business, Mongolian University of Life Sciences, 2018, p.207-211.
[7] B.A. McGregor, Variation in the softness and fibre curvature of cashmere, alpaca,
Calibrated CVS system captures the image containing color infor­ mohair and other rare animal fibres, The Journal of The Textile Institute 105 (6)
mation with millions of pixels, which enables further analysis for inner (2014) 597–608, https://doi.org/10.1080/00405000.2013.828448.
variation coefficients (CV) of color values, while it can be difficult to [8] C. Hu, H. Mei, H. Guo, J. Zhu, Color analysis of textile fibers by
microspectrophotometry, Forensic Chem. 18 (2020) 100221, https://doi.org/
obtain with a spectrophotometer. Fig. 11 demonstrated the CV of
10.1016/j.forc.2020.100221.
chroma indicators (a and b) was bigger than brightness indicator (L), [9] H.J. Swatland, Measurement of Veal Color by Fiber Optic Spectrophotometry,
and the CV of Lab values of darker cashmere is greater than lighter J. Food Sci. 50 (5) (1985) 1489–1490, https://doi.org/10.1111/jfds.1985.50.issue-
510.1111/j.1365-2621.1985.tb10507.x.
cashmere, which can guide the subsequent decolorization and dyeing
[10] S. Walbridge-Jones, Microspectrophotometry for textile fiber color measurement,
process of cashmere. Identification of Textile Fibers. (2009) 165–180.
[11] S.M. Goñi, V.O. Salvadori, Color measurement: comparison of colorimeter vs.
computer vision system, J. Food Meas. Charact. 11 (2) (2017) 538–547, https://
4. Conclusion doi.org/10.1007/s11694-016-9421-1.
[12] X. Cui, Y. Cai, J. Rodgers, V. Martin, M. Watson, An investigation into the intra-
Precise and inexpensive measurement of cashmere color is very sample variation in the color of cotton using image analysis, Text Res J 84 (2)
(2014) 214–222, https://doi.org/10.1177/0040517513490055.
important in determining its price and application. For this aim, a CVS
[13] T. Komatsu, S. Mohammadi, L.S. Busa, et al., Image analysis for a microfluidic
consisting of a digital camera, a controlled illumination environment, paper-based analytical device using the CIE L*a*b* color system, Analyst 141
and a software to process images was developed, which can be cali­ (2016) 6507–6509, https://doi.org/10.1039/c6an01409g.
brated using common spectrophotometer. This research proposed and [14] M. Tańska, D. Rotkiewicz, W. Kozirok, I. Konopka, Measurement of the geometrical
features and surface color of rapeseeds using digital image analysis, Food Res. Int.
implement a methodology for the determination of experimental con­ 38 (7) (2005) 741–750, https://doi.org/10.1016/j.foodres.2005.01.008.
ditions and color calibration. Proper experimental conditions including [15] A. Girolami, F. Napolitano, D. Faraone, et al., Measurement of meat color using a
temperature/humidity (20 ± 2 ℃, 65 ± 4 % RH), applied pressure (4 kg) computer vision system, Meat Sci 93 (2013) 111–118, https://doi.org/10.1016/j.
meatsci.2012.08.010, 2012/09/18.
and thickness (15 mm) were determined to capture the true color of [16] F. Pedreschi, J. León, D. Mery, P. Moyano, Development of a computer vision
cashmere, and then the proposed calibration model ensured the accu­ system to measure the color of potato chips, Food Res. Int. 39 (10) (2006)
racy of final testing results. Results obtained from the calibrated CVS and 1092–1098, https://doi.org/10.1016/j.foodres.2006.03.009.
[17] W. Castro, J. Oblitas, T. Chuquizuta, H. Avila-George, Application of image
spectrophotometer in this study have a coefficient of determination (R2) analysis to optimization of the bread-making process based on the acceptability of
0.93 or more, indicating that it is possible to accurately, inexpensively, the crust color, J. Cereal Sci. 74 (2017) 194–199, https://doi.org/10.1016/j.
objectively measure the cashmere fiber color. Besides, the categories of jcs.2017.02.002.
[18] H. Zhang, D. Li, Applications of computer vision techniques to cotton foreign
cashmere color can be determined based on the color parameter (L) matter inspection: A review, Comput Electron Agr 109 (2014) 59–70, https://doi.
values obtained by calibrated CVS, which can be used as an alternative org/10.1016/j.compag.2014.09.004.
method of manual assessment. [19] H. Manninen, M. Paakki, A. Hopia, R. Franzén, Measuring the green color of
vegetables from digital images using image analysis, LWT - Food Science and
Technology 63 (2) (2015) 1184–1190, https://doi.org/10.1016/j.
CRediT authorship contribution statement lwt.2015.04.005.
[20] B. Zhang, W. Huang, J. Li, C. Zhao, S. Fan, J. Wu, C. Liu, Principles, developments
and applications of computer vision for external quality inspection of fruits and
Chong Heng: Conceptualization, Methodology, Software, Investi­ vegetables: A review, Food Res. Int. 62 (2014) 326–343, https://doi.org/10.1016/
gation, Writing – original draft. Hua Shen: Validation, Formal analysis, j.foodres.2014.03.012.
Visualization. Fumei Wang: Validation, Formal analysis, Visualization. [21] A.I. AbdelGawad Saad*, Nazeer El-Bialee Internal quality assessment of tomato
fruits using image color analysis, AgricEngInt: CIGR Journal 18 (2016) 339–352.
Yufeng Chen: Resources, Writing – review & editing, Supervision, Data [22] K.-J. Lee, B.-W. Lee, Estimation of rice growth and nitrogen nutrition status using
curation. color digital camera image analysis, Eur. J. Agron. 48 (2013) 57–65, https://doi.
org/10.1016/j.eja.2013.02.011.
[23] A. Joiner, Tooth colour: a review of the literature, J. Dent. 32 (2004) 3–12, https://
Declaration of Competing Interest doi.org/10.1016/j.jdent.2003.10.013.
[24] F. Kiessling, The changing face of cancer diagnosis: From computational image
analysis to systems biology, Eur Radiol 28 (2018) 3160–3164, https://doi.org/
The authors declare that they have no known competing financial 10.1007/s00330-018-5347-9, 2018/03/01.
interests or personal relationships that could have appeared to influence [25] Y.Z.J. Yang, L. Zhuo, et al. Cheek region extraction method for face diagnosis of
Traditional Chinese Medicine. IEEE International Conference on Signal Processing.
the work reported in this paper.
2013, p. 1663-1667.
[26] Xingzheng Wang, David Zhang, A high quality color imaging system for
Acknowledgements computerized tongue image analysis, Expert Syst. Appl. 40 (15) (2013) 5854–5866,
https://doi.org/10.1016/j.eswa.2013.04.031.
[27] Di Wu, Da-Wen Sun, Colour measurements by computer vision for food quality
This work was supported by the National Natural Science Foundation control – A review, Trends Food Sci. Technol. 29 (1) (2013) 5–20, https://doi.org/
of China (No.51673036) and the Fundamental Research Funds for the 10.1016/j.tifs.2012.08.004.
Central Universities and Graduate Student Innovation Fund of Donghua

10
C. Heng et al. Measurement 185 (2021) 109991

[28] HuiYu Jiang XH, JunJiang Zhu, Jun Wan, JinBo Yao. Studies on the photofading of [31] Cheng-Jin Du, Da-Wen Sun, Recent developments in the applications of image
alizarin, the main component of madder. Dyes and Pigments 2021; 185: 1-9. Doi: processing techniques for food quality evaluation, Trends Food Sci. Technol. 15 (5)
10.1016/j.dyepig.2020.108940. (2004) 230–249, https://doi.org/10.1016/j.tifs.2003.10.006.
[29] Fernando Mendoza, Petr Dejmek, José M. Aguilera, Calibrated color measurements [32] H.I. Yoon, J.W. Bae, J.M. Park, Y.S. Chun, M.A. Kim, M. Kim, A study on possibility
of agricultural foods using image analysis, Postharvest Biol. Technol. 41 (3) (2006) of clinical application for color measurements of shade guides using an intraoral
285–295, https://doi.org/10.1016/j.postharvbio.2006.04.004. digital scanner, Journal of Prosthodontics 27 (2018) 670–675, https://doi.org/
[30] Kit L. Yam, Spyridon E. Papadakis, A simple digital imaging method for measuring 10.1111/jopr.12559, 2016/11/07.
and analyzing color of food surfaces, J. Food Eng. 61 (1) (2004) 137–142, https:// [33] Vilbett Briones, Josè M. Aguilera, Image analysis of changes in surface color of
doi.org/10.1016/S0260-8774(03)00195-X. chocolate, Food Res. Int. 38 (1) (2005) 87–94, https://doi.org/10.1016/j.
foodres.2004.09.002.

11

You might also like