An investigation into the intra-sample variation in the color of cotton using image analysis

Abstract

The colorimeter method is widely used to identify the color and grade of a cotton sample, but this method does not give information about the variation of color within the sample. We have conducted an investigation using an image analysis method to study the intra-sample distribution and variation in cotton color. High-resolution color images of cotton samples were obtained using a color scanner. For each image the Rd and +b values of the image and specific sub-areas within it were computed to analyze the color distribution. Intra- and inter-sample variations in Rd and +b values and color grade were compared with those obtained by the HVI system. The results show that for the same cotton substrate, although the variations among the color grades of different samples (replicates) may be very small, intra-sample color variations are evident, the colors of the sub-areas being distributed over a range of areas of the cotton color chart. In other words, cotton samples of the same overall color grade may exhibit different intra-sample color variations and distribution. A description of the intra-sample color variations and distributions obtainable using an image analysis method may allow a more comprehensive evaluation of color characteristics on cotton.

Keywords

Cotton quality measurement color distribution image analysis

In the classification used in the cotton trade color is one of the most important properties assessed. The color of the cotton can provide an indication of the quality of the textile products produced from it.¹ The most commonly used method for measuring the color grade of cotton is the colorimeter. Modern colorimeters have been in use since the early 1950s and assess the color of a cotton sample based on its Hunter Lab reflectance (Rd) and yellowness (+b) values.² A look-up table of grades has been constructed from a series of Rd and +b pairs, and from this table the grade of a cotton sample can be determined once its Rd and +b values are known.

There have been a variety of studies conducted on the measurement of the color grade of cotton.^3–8 Past research has studied the relationship between color grade as reported by cotton classifiers and the color properties determined by the early disc colorimeters during the development of cotton color standards.³ A major focus of previous studies has been on the development of methods or algorithms for improving the measurement of the overall color grade of a cotton sample. These studies have included the use of modern techniques such as fuzzy logic to enhance analysis of the color information,⁴ as well as applying artificial neural networks to evaluate the impact of impurities such as trash particles.⁵ In recent years efforts have been made to use CIE color parameters or to use color spectrophotometers to develop traceable color standards for cotton grading.^6–8 However, any discussion in the literature of the characteristics of intra-sample (within-sample) distribution, and variation between single measurements, has been lacking.

Although the overall color of cotton is currently determined by the colorimeter method, this does not provide information about color distribution and variation within a single sample (intra-sample variation). Since cotton color is the result of naturally occurring pigments it is a collective property of individual fibers, and the color of cotton shows a variation even within the same sample. This is analogous to cotton fiber length: a cotton sample has an average fiber length, but the fiber length has its own distribution and variation within a sample.⁹

In general, uniform fiber properties are desirable for consistent textile processing and product quality. It is well known that in selecting cotton bales for a spinning laydown, it is not only the average value of the properties of the bales (such as length, denier, or color) that is important, but also that the variation in the values of each should lie within a narrow range. The variation in these properties within the bale is also important, but unfortunately these variations are more difficult to measure. Since color provides an indication of the overall physical properties of a cotton sample, variation in color may suggest variability in other physical properties that influence processability and product quality.

We conducted an investigation using image analysis to examine the data from color images scanned from cotton samples. Instead of examining the overall color grade, our research focused on intra-sample color distribution and variation. For an individual cotton sample, the intra-sample distribution and variation can be obtained from the color image of the sample; the image contains color information within the measurement window upon which the individual sample is placed. This investigation can provide new insights into the color characteristics of cotton, and we expect the results to give a better and more rapid assessment of cotton quality.

Materials and method

An Epson Perfection V500 scanner was used to acquire the color images of the cotton samples. In addition to the scanned color images, a Uster HVI–1000 (Uster Technologies, Knoxville, TN) was used for measuring the Rd and +b values of samples and color grades. Color values reported by the HVI were compared with the results of image analysis.

During the scanning process, the sample was placed against a scanning window 114 mm × 114 mm in size. A 5 kg weight was applied over the sample. The scan resolution was 400 dpi, and images were stored in 48-bit sRGB format. When processing each image, an 88.9 mm × 88.9 mm area was selected to simulate the size of the HVI color and trash measurement window, which had a similar area. The sRGB values of each pixel were used as original inputs to compute Rd and +b values and then the color grades. When computing the Rd and +b values, the images were converted from the original sRGB color space to CIE XYZ space, and the Rd and +b values were calculated using the following equations for Illuminant C and 2° standard observer.¹⁰ Equations (1) give a definition of these in terms of the CIE tristimulus values, X, Y and Z.

\begin{matrix} f (Y) = 0.51 (21 + 0.2 Y) / (1 + 0.2 Y), \\ Rd = Y, and \\ + b = K_{b} f (Y) (Y / Y_{n} - Z / Z_{n}), \end{matrix}

(1)

where K_b = 70.00, Y_n = 100.00, and Z_n = 118.11.

After Rd and +b values have been obtained, the color grade of a cotton sample can be determined using a look-up table provided by the Cotton Program of USDA, Agriculture Marketing Service (AMS). The look-up table consists of a series of Rd and +b data pairs, together with the corresponding cotton color grades. Matlab programs were developed to carry out the color value computations and locate the grades from the look-up table. This process is illustrated in Figure 1.

Figure 1.

Method for computing color grade from scanned images.

In order to study the intra-sample variation in the Rd and +b values of a sample, the 88.9 mm × 88.9 mm area selected from the image of the sample was divided into sub-areas, and the sub-area size was decreased successively from 12.7 mm × 12.7 mm and 6.35 mm × 6.35 mm down to a single pixel. The color values and grades of each sub-area were computed using the process described.

Figure 2 shows an example of how an image is divided into a total of 49 sub-areas of 12.7 mm × 12.7 mm. The distributions and variations can then be investigated from the data in the sub-areas.

Figure 2.

Division of the image into sub-areas and their Rd and +b values and grades.

Using the method described above we tested different batches of cotton samples with a wide range of color values. Although there has been no report that cotton color measurement is influenced by temperature or relative humidity, as a precaution all cotton samples were kept under standard condition for at least 24 h before testing, and all measurements were carried out at 21 ± 1℃ and 65 ± 2% RH.

Results and discussion

Comparison of results between image analysis and HVI

Firstly, we examined whether the Rd and +b values obtained by image analysis would provide a linear relationship with the HVI results. Figure 5(a) and 5(b) show comparisons of Rd and +b values from scanned images and HVI–1000 results from 110 samples. It can be seen that although scanned Rd and +b gave values slightly different from those by HVI, the relationship between them was highly linear. In both Figures the correlation coefficient is 0.99. These comparisons confirm that the color values from the scanner corresponded closely with those by HVI color measurement. The small differences may include measurement variations and differences between individual color measurement devices. Although the absolute values may be different, the high degree of linearity confirms that the results can be calibrated using standard materials, such as USDA or NIST color tiles. Although matching the results from image analysis and HVI measurement is not central to this paper, this is a very important aspect since the colorimeter method used in the HVI is widely used in determining cotton color grades in the cotton industry. This linear relationship therefore confirms that the color values and color distribution obtained from the scanned images can be used to study the corresponding results obtained by the colorimeter method.

Figure 3.

Changes in Rd and +b intra-sample variations with respect to sub-area size.

Figure 4.

Rd and +b histograms of sub-areas with respect to sub-area size. (a) Rd and (b) +b.

The results from color image analysis and HVI color measurement were both used to study inter-sample variation in Rd and +b values, based on the overall color variation between five samples taken from different areas of the same cotton bale. Table 1 shows the average inter-sample coefficient of variation (CV_inter) of Rd and +b values between 110 samples. Within the same cotton bale the color CVs from the 88.9 mm × 88.9 mm window were very small. From Table 1, it can be seen that results from color image analysis and HVI color measurement showed a greater variation in +b than in Rd.

Table 1.

Average inter-sample color variations in Rd and +b.

Methods	Rd CV_inter	+b CV_inter
HVI	0.65%	1.92%
Scanner	0.82%	2.50%

CV: coefficient of variation.

The image analysis results indicate that intra-sample Rd and +b variations were much higher than the inter-sample variations (Figure 3). Intra-sample variations were evaluated by comparing the coefficient of variation (CV_intra) computed from the Rd and +b values of the sub-areas of the same sample. Similar to inter-sample variations, the results also show that the +b intra-sample variation was greater than the Rd intra-sample variation. In addition, the results show that the calculated intra-sample variations in Rd and +b increased with decreasing sub-area size, reaching a maximum when the sub-area size reached pixel level. This can be explained from the computation of the color values of the sub-areas: the Rd and +b values of a sub-area are mean values computed using the values of individual pixels within the sub-area. If the sub-area size is bigger, the average values are then generated from a greater number of individual pixels, consequently reducing the intra-sample variation of the entire sample image. In the extreme case of the sub-area equaling the image itself, the intra-sample variation cannot then be observed. The Rd and +b histograms of a sample (Figure 4(a) and 4(b)) also support the fact that, in general, if the sub-area size is larger, the Rd and +b values are concentrated within a smaller range, resulting in less intra-sample variation.

The spatial characteristics of intra-sample color variation on the color chart

Comparisons of the data in Table 1 and Figure 3 show that the intra-sample color variation was much greater than the inter-sample color variation. To obtain more a straightforward understanding of the significance of intra-sample color variation, color grade charts obtained from USDA AMS were used to display the distribution of color grades of the sub-areas of a cotton sample. For a sub-area size of 6.35 mm × 6.35 mm a measurement window of 88.9 mm × 88.9 mm has 196 sub-areas. Figures 6 –8 show intra-sample and inter-sample color grade distributions for three different cottons. These three cottons were chosen because of their clearly different color grade. The figures were generated by plotting the Rd and +b values of the sub-areas on the cotton color grade chart, compared respectively to the overall grades obtained from five different samples of the same cotton. All the inter-sample plots of these three cottons indicate that the color grades from different samples of the same cotton are very uniform (small inter-sample variation), as shown in Figures 6(b), 7(b), and 8(b) (closely packed together).

Figure 5.

Comparison between color values from scanned images and those from HVI. (a) Rd and (b) +b.

Figure 6.

Intra-sample and inter-sample color grade distributions of cotton #1. (a) Color grades of sub-areas and (b) Color grade of five samples of cotton #1.

Figure 7.

Intra-sample and inter-sample color grade distributions of cotton #2. (a) Color grades of sub-areas and (b) Color grade of five samples of Cotton #2.

Figure 8.

Intra-sample and inter-sample color grade distributions of cotton #3. (a) Color grades of sub-areas and (b) Color grade of five samples of Cotton #3.

Although the inter-sample color variation was usually very small, the results further revealed that the color grades of the sub-areas were not uniform. Even for the color image of the same cotton, color grades were distributed around an area on the color chart, instead of being concentrated only at a single location. The figures show that intra-sample color variations were significant. On the color grade charts, all intra-sample variations covered a region containing both higher and lower grades (Figures 6(a), 7(a), and 8(a)), and the intra-sample color grade region was much larger than the inter-sample color grade region plotted in the corresponding inter-sample color chart, as shown in Figures 6(b), 7(b), and 8(b).

The influence of intra-sample color variations

The current colorimeter method provides only one color grade, based on the overall Rd and +b values of the cotton sample; and the intra-sample color variation and distribution cannot be determined. As shown in Figures 6–8 the large and complex intra-sample color variations indicate that one overall color grade is apparently not sufficient to characterize comprehensively the color properties of a cotton sample. The image analysis method, on the other hand, is able to examine the color distribution within one image of a sample. A further investigation of cotton color variation and distribution properties indicates that, even when two cotton samples have the same overall color grade, the different variations could lead to differing color distribution characteristics between them.

Figure 9 displays two cottons that have the same overall color grade; their overall Rd and +b values are very close. Meanwhile the intra-sample color variation of their Rd and +b are different. The Rd and +b values and color grades are listed in Table 2.

Figure 9.

Color images of two cotton samples of the same overall color grade. (a) Sample #4 and (b) Sample #5.

Table 2.

Overall Rd and +b values and grades, and intra-sample color variations of samples #4 and #5

Sample	Rd	CV Rd	+b	CV +b	Grade
#4	65.51	5.1%	9.21	15.0%	52-1
#5	65.50	4.8%	9.18	11.5%	52-1

CV: coefficient of variation.

The grade of each 6.35 mm × 6.35 mm sub-area was determined from the color grade look-up table, and the distribution of color grades of the 88.9 mm² measurement window was then generated. Since the grade distribution is jointly determined by pairs of Rd and +b values, the distribution can provide a more useful observation of color properties than the Rd or +b value alone. By examining these distributions (Figure 10(a) and 10(b)), it can be seen that the overall grade of these two samples is grade 52, but only 30–40% of the sub-areas are actually grade 52, and the remaining sub-areas have a grade either better or worse than this. Moreover, though the overall color grades of the above two samples are the same, when we look at the color grades of the 196 6.35 mm × 6.35 mm sub-areas of the 88.9 mm × 88.9 mm image, their grade distributions are different (Figure 10). This difference indicates that it might be important and helpful to develop an additional parameter which describes the intra-sample color distribution and assesses the extent of the variation, in order to evaluate more comprehensively the color properties of a cotton sample.

Figure 10.

Color grade distributions of the two samples in Figure 9. (a) Sample #4 and (b) Sample #5.

Further discussion

Although this preliminary study has demonstrated that the image analysis method can measure the cotton color distribution and hence calculate statistical parameters such as average and standard deviation, a number of issues need to be further studied. One of these is the effect of trash on cotton color measurement. Impurities or trash exists in commercial cottons, and the trash particles show as dark spots. Based on USDA classification data for more than 14 million bales, the average area of the dark spots is 0.34% of the sample surface area. Research has shown that the trash does not have a significant influence on cotton color grade.² On the other hand, the trash may have a more significant influence on cotton color distribution.

Another issue is the unevenness or roughness of the sample surface, which may influence the color distribution measurement. Applying pressure on the sample can reduce the roughness, but cannot eliminate it. We are therefore working on ways of digitally reducing the influence of trash and roughness of the sample surface. For the commercial application of image analysis to measurement of cotton color, a proper calibration method needs to be developed to ensure the consistency of the measurement, especially when using different instruments.

To our knowledge, the present study is the first to investigate the cotton color distribution. This new tool may lead to further research to relate cotton color variation to variations in other fiber properties and product quality, particularly uniformity of dyeing.

Conclusions

The colorimeter method provides the Rd and +b color values of a cotton sample, from which the color grade of the cotton can be assessed. The result does not however include information about the color distribution and variation within the sample. We have used a computer image analysis method to investigate the intra-sample distribution and variation in cotton color. The area of the scanned image of each sample was divided into sub-areas to compute the Rd and +b values, and its color grade. The color values of the sub-areas were then used for analyzing the intra-sample color distribution. Rd and +b values from image analysis showed a strong linear relationship with the results from HVI color measurement. For the same cotton, although the inter-sample color variation was very small, the intra-sample color variations were significantly greater.

The intra-sample variations in Rd and +b values increased with decreasing sub-area size. The intra-sample color had a wide distribution spanning large areas of Rd and +b on the cotton color grade chart. Cottons with the same overall Rd and +b and color grade may have different intra-sample color variation and distribution. This indicates that at least one further parameter characterizing color variation and distribution is needed in addition to the overall color values and grade.

The results show that the computer image analysis method can effectively be used to measure cotton color distribution and variation.

Footnotes

Disclaimer

Names of companies or commercial products are given solely for the purpose of providing specific information; their mention does not imply recommendation or endorsement by the US Department of Agriculture over others not mentioned.

Funding

This work was supported by Cotton Incorporated.

Acknowledgements

We wish to thank USDA AMS for providing the samples and the color grade look-up table and chart.

References

Duckett

Zapletalova

Cheng

Ghorashi

Watson

. Color grading of cotton part I: spectral and color image analysis. Textile Res J 1999; 69(11): 876–886.

Thomasson

. Foreign matter effects on cotton color measurement: determination and correction. Trans Am Soc Agric Biol Engineers 1993; 36(3): 663–669.

Nickerson

. Color measurements of standards for grades of cotton. Textile Res J 1946; 16(9): 441–449.

Dale

Huang

Watson

. Cotton color classification by fuzzy logic. Textile Res J 2002; 72(6): 504–509.

Kang

Kim

. Objective evaluation of the trash and color of raw cotton by image processing and neural network. Textile Res J 2002; 72(9): 776–782.

Shofner F, Watson M, Zhang Y, Lee S and Shofner K. Moving to CIE color, traceably! In: Proceedings of the 2006 Beltwide Cotton Conference, San Antonio, TX, 2006. pp.1921–1925.

Thibodeaux D, Campbell J and Knowlton J. Development of NIST - traceable HVI color measurements. In: Proceedings of the 2005 Beltwide Cotton Conference, New Orleans, LA, 2005. p.2443.

Rodgers

Thibodeaux

Cui

Martin

Watson

Knowlton

. Instrumental and operational impacts on spectrophotometer color measurements. J Cotton Sci 2008; 12(3): 287–297.

Xiao

Rodgers

Cai

Belmasrour

Pang

. Obtaining cotton fiber length distribution from the beard test method part I – theoritcal distribution related the beard method. J Cotton Res 2009; 13: 265–273.

10.

Hunter Lab. Insight on color: R_dab color scale. HunterLab Appl Notes 2008; 8(4): 1–2.