Skip to main content
ARS Home » Southeast Area » New Orleans, Louisiana » Southern Regional Research Center » Cotton Structure and Quality Research » Research » Publications at this Location » Publication #259673
Title: Fabric wrinkle characterization and classification using modified wavelet coefficients and optimized support-vector-machine classifier

Author
item SUN, JINGJING - University Of Texas
item YAO, MING - University Of Texas
item XU, BUGAO - University Of Texas
item Bel, Patricia

Submitted to: Textile Research Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/26/2011
Publication Date: 6/1/2011
Citation: Sun, J., Yao, M., Xu, B., and Bel, P., 2011. Fabric wrinkle characterization and classification using modified wavelet coefficients and optimized support-vector-machine classifier. Textile Research Journal. 81(9):902-913.

Interpretive Summary: Wrinkling is caused by wearing or laundering procedures and is a vital performance characteristic of fabric. Conventionally, fabric wrinkling is evaluated by visual examination of fabric samples performed by trained experts in accordance to the standard wrinkling samples (e.g., the AATCC Smoothness Appearance (SA) replicas). This paper presents a novel wrinkle evaluation method that uses modified wavelet coefficients and an optimized support-vector-machine (SVM) classification scheme to characterize and classify wrinkle appearance of fabric. The effectiveness of this evaluation method was tested by 300 images of five selected fabrics with different fiber contents, weave structures, colors and laundering cycles. A number of the parameters that characterize wrinkle directionality, hardness, density, and contrast can be defined based on the modified coefficients. It was demonstrated that these parameters are useful in differentiating oriented or random, hard or soft wrinkles and effective in discriminating fabric images in terms of the AATCC SA standards when used as inputs to SVM classifiers. A cross-validated test shows the RBF SVM classifier and the linear SVM classifier can have, respectively, 78% and 75% successful rates in comparison to the visual ratings. The presented method provides an efficient way for both the detailed characterization of wrinkling features and the overall grading of smoothness appearance. The parameters can provide detailed information useful for product improvements and quality control for textile, appliance, and detergent manufactures that study wrinkling behaviors of fabrics.

Technical Abstract: This paper presents a novel wrinkle evaluation method that uses modified wavelet coefficients and an optimized support-vector-machine (SVM) classification scheme to characterize and classify wrinkle appearance of fabric. Fabric images were decomposed with the wavelet transform (WT), and five parameters were defined based on the modified wavelet coefficients to describe features such as wrinkle directionality, hardness, density, and contrast. These parameters were also input into an optimized SVM classifier to obtain overall wrinkle grading for the evaluated sample in accordance with the standard AATCC smoothness appearance (SA) replicas. The SVM classifiers based on linear kernel and radial basis function (RBF) kernel were used in this study. The effectiveness of this evaluation method was tested by 300 images of five selected fabrics with different fiber contents, weave structures, colors and laundering cycles. The cross-validation test of the classification indicated that more than 75% of these diversified samples could be recognized correctly. The parameters can provide detailed information useful for product improvements and quality control for textile, appliance, and detergent manufactures that study wrinkling behaviors of fabrics.