Assessing structural changes in knits during processing

Assessing structural changes in knits during processing

Abou-Iiana, M


An automatic fabric evaluation system is developed to analyze knit structures and objectively evaluate fabric properties. Fabric images are captured with a CCD camera, and digital images are processed by histogram equalization, binary morphological operators, and pattern recognition. Six knitted fabrics are evaluated at different phases of manufacturing. Fabric construction parameters such as courses and wales per unit length, fabric cover, and areal density are measured and evaluated using the new approach and manual methods. Structural changes that occur in the fabrics at different levels of relaxation are also documented. Our image capturing and analysis system is capable of on-line control of knit structures’ spatial characteristics before and after wet treatments.

For years, knitting has been considered more an art than a science. Many attempts have been made over the past century to quantify the characteristics of knitted fabrics [2, 3, 6, 10, 11]. The key to understanding a knitted structure lies within its basic element, the single knitted loop. The length of yarn knitted into a single loop will determine such overall fabric qualities as hand, comfort, weight, extensibility, finished size, cover factor, and most importantly, fabric dimensional stability [8]. Therefore, control of the characteristics of a fabric in order to meet certain performance criteria can be achieved through control of individual knitted loops. The problem then arises of how to determine that a knitted loop is the correct size and shape for a given set of fabric properties.

The answer lies in the ability to objectively measure the knitted loop size/shape during processing. Once the loop shape in a fabric is measured, then loops of that size/shape can be correlated with specific properties of that fabric. In this work, we employ image analysis and processing to solve this age-old problem. Using our proposed technique during fabric processing, we can characterize a loop with a good degree of accuracy (as opposed to previous inaccurate techniques of measuring course spacing). Digital imaging not only provides an accurate method for characterizing loop shape but also a means for comparing loop shape to the desired shape chosen for particular fabric performance characteristics.

Attempts to specify the dimensional properties of a knitted fabric in terms of length or width parameters (courses and wales per unit length) are subject to inaccuracy because the knit has inherent stretch at low loads and poor recovery. Thus, determining the “relaxed” state at which to measure these dimensional parameters is difficult, and the error rate is high. Nevertheless, the construction of a fabric today is still frequently described in terms of courses and wales per unit length. The use of this approximate and averaged parameter for specifying the tightness of a knitted construction is responsible (directly or indirectly) for many of the problems associated with the dimensional control of knitted structures.

In an industrial setting, techniques of counting the courses and wales per unit length or unraveling the fabric to determine stitch length are subject to human error and are time consuming. An automatic structure analysis and objective evaluation of knit structures using image analysis techniques can determine fabric construction parameters while eliminating human subjectivity.

Review of Prior Work

Currently, image analysis tools are being used to characterize the structural parameters of textiles during processing. Emery and Buchanan [1] have described a system for automatic inspection of yarn packages of openend polyester/cotton blends using thirty-six image frames to inspect the packages for different flaws. Kang et al. [5] developed a system for objective evaluation of woven fabric using image analysis techniques. Later, Xu introduced a different approach for identifying woven fabrics, pattern recognition, and yarn count with fast Fourier transform [15].

To evaluate knit structures, Jensen et al. [4] developed a system to evaluate knit fabric surfaces for fuzz and pills with digital image analysis. They used Fourier analysis to filter the knitted stitch background from the fuzz and pills on the textile with an automatically generated Fourier mask. To examine the reflective properties of fine knit fabrics in the leg section of pantyhose, Morooka et al. [7] used an adjustable angle optical imaging device. Their results showed the effects of respective yarn structures on the reflectiveness of the cloth.

Wood [12] used image-processing methods to measure the pile texture of carpet. A Fourier power spectrum provided a useful description of the spatial frequency content in a digital image, and in particular the coarseness of any texture present. In more detailed studies, Wood [13, 14] described techniques for measuring the appearance characteristics of carpets with digital image analysis and the key appearance parameters that are relevant to carpets. He used a gray-scale image analysis to measure texture periodicity and tuft spatial distribution. The system revealed that mechanical wear increases the average tuft-to-tuft distance, decreases spatial density, and promotes a tendency toward randomness.

System Configuration, Image Analysis, and Image Processing

To develop a method for automatic structure analysis and objective evaluation of knitting parameters, we have devised a system consisting of a CCD camera as the input device, a lighting device, and a PC as the image analyzing device. This system allows quick, precise data acquisition and easy statistical analysis of the results. Because of the inherently flexible nature of knit structures, it was crucial to select software appropriate for knits that provides all the necessary image processing and analysis. We made a thorough survey of software options that provide a scripting or programming language for image processing of knit structures. We decided to use imaq from Labview(R) because of its image analysis abilities and pattern recognition interface.

Initial image capturing trials with knit structures indicated that specimen preparation and alignment are critical elements for image clarity. Determination of orthogonal spacing of courses and wales is associated with fabric alignment. If the fabric is skewed, sheared, or rotated with respect to the camera frame, then the courses and wales are similarly not aligned with the two camera orthogonal directions. A simple rotation of the fabric image is not always sufficient-sometimes due to shearing or skewing, the course and wale directions are not perpendicular. After several trials with different fabric samples made from different yarn sizes and knitted on machines with different gauges, we were able to adjust the image capturing and retrieve useful data for analysis.

The main steps for image processing in this work are equalization (to balance the overall gray scale in the image), followed by a threshold operation to identify only those features with grayscale values below a certain value, and then an erosion/dilation operation to remove small artifacts from the image. At this point, the loop appears in the image, and a pattern recognition module is applied to that image. This module searches the image for similar shapes and reports key information about each pattern, such as location in the image (x and y coordinates), total number of pixels, smallest circumscribable rectangle, etc. Figure 1 illustrates a single loop and the change in the image as a result of the various image operations described above.

Experimental Work

We used two sets of knitted fabric samples in this study. The first set was knitted on a Bentley TM machine at 10 needles per inch, 396 total needles, and equipped with positive feed. National Textiles in Greensboro, NC, provided the second set of samples, which were a series of knits showing the fabric after each manufacturing step. To capture the images of these knitted fabrics, samples were laid on a flat surface with minimum tension to remove wrinkles. The samples edges were prevented from curling by adhering the edges to a flat surface.


Courses and wales per unit length were determined from pattern recognition of single loops, defined as repeat units and teaching the software their detailed features. The search mode was defined by translation parameters, while disallowing any rotational parameters (i.e., assuming that loops will not have a major in-plane angle rotation). Raw data for loop center locations in two dimensions were obtained and sorted by X- and Y-axis values to demonstrate data accuracy. We determined the average yarn spacing between the wales by averaging the difference between the X-axis values, and the average yarn spacing between the courses by averaging the difference between the Y-axis values. From this, we found the number of wales and courses per unit length. Each fabric sample was imaged with a scale included so that pixel dimensions could be converted to physical dimensions of the fabric. An edge detector process was applied to the scale, followed by a caliper function to calculate the number of pixels per inch in each image.


Fabric cover factor is one of the parameters sought by our system. The cover factor can be evaluated by considering the planar projection of the fabric onto its inherent course-wale plane. The amount of fibers projected onto a given area proportioned to the area under consideration is the cover factor. Thus, using image analysis, it is possible to determine fabric cover by choosing a gray level corresponding to the lightest fiber and applying a threshold filter to everything with a darker shade. This would capture all of the visible fibers in the image. Then the cover factor is calculated as the ratio of the number of black pixels to the total number of pixels in the image.

We used this method with both lighting from above the fabric (reflective) and lighting from below (transmissive). Results were different for the two methods. From a theoretical point of view, the cover factor [9] for plain knitted structures is

K = K^sub s^D/L , (1)

where K = cover factor, L = loop length, D = yarn diameter, and K^sub s^ = fabric constant. This assumes a cylindrical yarn (no hispid features) and knowledge of the fabric constant. Comparing the theoretical results from Munden’s model with the experimental image analysis data, we observed (not unexpected) differences.

Comparing the two lighting techniques with Munden’s model, the below-surface lighting (transmission) had the strongest correlation. To further investigate the structural changes during relaxation and their effect on the image, structural images were taken at dry, wet, and finished relaxation stages.


Fabric areal density (sometimes called fabric weight) is the total weight of yarn forming the loops per unit area. Knowing the number of courses and wales per unit length, yarn count, and stitch length, we can determine the fabric weight using the following relationship:

where W = fabric weight in oz/yd2^sup 2^, cpi = courses per inch obtained from the automatic structure analysis, wpi = wales per inch obtained from the automatic structure analysis, L = stitch length in inches, and N^sub ec^ = yarn English count.

The areal density of the knit will change as the fabric relaxes. Loop spacing decreases during relaxation, but stitch length does not, so areal density increases. Using the automatic structure analysis, we determined fabric parameters (courses and wales per unit length) at different stages of relaxation. We ran a series of experiments in which we measured fabric weight experimentally using ASTM D 3775 (1999) option C and compared these data to the data obtained from Equation 2, where the courses and wales per unit length were evaluated with image analysis.

Results and Discussion

To compare results from our automatic fabric evaluation system to experimental values, we conducted manual analyses of selected fabric samples. Table I provides the process code to describe the different processing steps, after which the fabric was extracted for evaluation. Table II shows the yarn type used in each of the fabrics in this study. Fabrics are thus referenced by the fabric code (Table II) and the processing step (Table I).

We measured the fabric count optically with a pick glass and determined the cover factor using Equation 1. The results from the automatic structure analysis system and the experimental values are compared in Figures 2, 3, and 4.

Figure 2 compares the measured course and wale spacings with image analysis and manual techniques. With few exceptions (notably fabrics after the drying operation), the manual method provides a higher count than the image analysis method. Compared to the manual system, the error in the image analysis system averages approximately 7% for a given fabric system.

Figure 3 compares cover factors determined by image analysis with those calculated by Munden’s equation. We used reflective light to acquire images and calculate the cover factors in Figures 3a-f and both light systems in Figure 3g. The correlation between results with reflective light images and the Munden equation was not good. On the other hand, transmissive light images matched well with Munden’s model, with an average error value less than 5%. The discrepancy with reflective light can be associated with several possible reasons, including overlap of knitted loops, effect of light diffraction through the spaces between the loops, and assumptions of idealization associated with Munden’s model used in calculating the cover factor. The evaluation of which measurement is a more accurate interpretation of the fabric cover factor is left as an exercise for the reader.

Figure 4 compares the fabric areal densities from manual measurements to results from image analysis. Curiously, there is no obvious trend-the image analysis under-predicts for fabric 1721 and over-predicts for fabrics 1725 and 1856. Overall, the error rate is relatively low (10-12%).

In general, we observed that most fabric parameters measured with our evaluation system are slightly different from those evaluated manually. This can be due to the fact that the machine vision application searches for a loop template in the image and sometimes ignores some of stitches when they do not match the template to a certain degree of accuracy.


We have developed a system to measure knitted fabric parameters with image analysis techniques. Digital images are captured by a ccd camera and preprocessed using histogram equalization, binary morphological operators, and pattern recognition. This technique can evaluate courses and wales per unit length, fabric cover, and weight per unit area.

We have tested six fabric samples after various processes and at different relaxation conditions and documented the structural changes that occur in these fabrics with the new approach. We have used manual techniques to evaluate fabric structural parameters and compared the results with those from the automatic fabric evaluation system. There is a good level of accuracy for both approaches.

Literature Cited

1. Emery, N. B., and Buchanan, D. R., Yarn Package Inspection with Computer Vision and Robotics, Textile Res. J. 58, 605-614 (1988).

2. Heap, S. A., Greenwood, P. F., Leah, R. D., Eaton, J. T., Stevens, J. C., and Keher, P., Prediction of Finished Weight and Shrinkage of Cotton Knits-The Starfish Project, Textile Res. J. 53, 109-119 (1983).

3. Hepworth, B., Hepworth, K., and Leaf, G. A. V., The Plain-Knitted Structure, Textile Res. J. 40, 863 (1970).

4. Jensen, K. L., and Carstensen, J. M., Fuzz and Pills Evaluated on Knitted Textiles by Image Analysis, Textile Res. J. 72, 34-38 (2002).

5. Kang, T. J., Choi, S. H., Kim, S. M., and Oh, K. W., Automatic Structure Analysis and Objective Evaluation of Woven Fabric Using Image Analysis, Textile Res. J. 71, 261-270 (2001).

6. Knapton, J. J. F., Richards, S., and Fong, W., The Dimensional Properties of Knitted Wool Fabric, III: The Plain-Knit Structure in Machine-Washing and Tumble-Drying, Textile Res. J. 40, 543-553 (1970).

7. Morooka, H., Wakashima, K., and Matsumoto, Y., Reflective Properties of Pantyhose Cloth as Analyzed by an Image Analysis System, Textile Res. J. 69, 68-74 (1999).

8. Munden, D. L., The Geometry and Dimensional Properties of Plain-Knitted Structure, J. Textile Inst. 50, T448-T464 (1959).

9. Postle, R., Doctoral thesis, Leeds University, U.K., 1965.

10. Shanahan, W. J., and Postle, R., A Theoretical Analysis of the Plain-Knitted Structure, Textile Res. J. 40, 656-665 (1970).

11. Smirfitt, J. A., Worsted 1X1 Rib Fabrics, Part I: Dimensional Properties, J. Textile Inst. 56, T248-T298 (1965).

12. Wood, E. J., Objective Measurement of Carpet Appearance by Image Analysis, Part I: Principles and Methodology, Melliand Textilber. 77, 452-459, E99-102 (1996).

13. Wood, E. J., Objective Measurement of Carpet Appearance by Image Analysis, Part II: Applications, Melliand Textilber. 77, 845-850, E182-184 (1996).

14. Wood, E. J., Applying Fourier and Associated Transforms to Pattern Characterization in Textiles, Textile Res. J. 60, 212-220 (1990).

15. Xu, B., Identifying Fabric Structure with Fast Fourier Transform Techniques, Textile Res. J. 66, 496-506 (1996).

Manuscript received June 18, 2002; accepted August 26, 2002.


School of Textiles and Materials Technology, Philadelphia University, Philadelphia, Pennsylvania 19144, U.S.A.


Department of Textile Engineering, Auburn University, Auburn, Alabama 36849, U.S.A.

Copyright Textile Research Institute Jun 2003

Provided by ProQuest Information and Learning Company. All rights Reserved