ARS | Publication request: Unsupervised Defect Detection in Size-Free Woven Fabrics using Wavelets and Image Morphology

Submitted to: International Conference on Composites Engineering Proceedings
Publication Type: Proceedings
Publication Acceptance Date: June 1, 2007
Publication Date: July 15, 2007
Citation: Sawhney, A.P., Hossain, I., Singh, K., Condon, B.D., Pang, S., Sachinvala, N.D. 2007. Unsupervised Defect Detection in Size-Free Woven Fabrics using Wavelets and Image Morphology. International Conference on Composites Engineering Proceedings. p. 831-832.

Interpretive Summary: Cotton woven fabrics are produced by using two sets of yarns, namely the warp and weft or filling. Traditionally, the warp yarns almost always are sized or coated with an adhesive to enable the yarns to withstand the rigorous stresses and abrasion caused by the weaving process or mechanism and thus to achieve efficient weaving of defect-free fabrics. However, the process of warp sizing is complex, costly and, more importantly, environmentally sensitive. Obviously, the textile industry wants to reduce and preferably eliminate warp sizing, which, as well, would eliminate another tedious, costly and ecologically-unsafe process of fabric desizing. The ARS-USDA scientists at SRRC are putting their best efforts to weave a cotton fabric from a size-free warp on a modern, high-speed weaving machine operating under mill-like conditions. Although hundreds of yards of fabric have been woven with a size-free warp without any yarn breakage or failure, the fabric quality is still unsatisfactory. Minor ball-like defects, most likely caused by intensive abrasion of the warp yarns, appear on the fabric surface. In order to try to eliminate these yarn/fabric defects, the first step is to efficiently identify and quantify these defects, so that any progress made in this regard can be properly evaluated. This paper presents a novel approach to do so by using digital image processing (DIP) technology.

Technical Abstract: In this paper, automated, objective detection and (possibly) quantification of fabric defects, using Digital Image Processing (DIP) techniques in conjunction with a high-resolution digital camera and a computer, are presented. This novel method of so-called unsupervised detection of fabric defects utilizes techniques of wavelet decomposition and reconstruction, histogram analysis, and binary morphological operators. For simplicity, only the soft ball-type defects, which specifically were encountered in size-free weaving trials, are the candidates for the subject identification and quantification method. However, the method may be extended to conventional types of fabric defects, as well. The identification technique exploits the fact that if a digital image of a fabric defect is captured, the characteristics of the image will deviate from those of the normal (defect-free) fabric image. This exhibits a lot of linear features in an inherently perpendicular orientation (due to the presence of orthogonal yarns in the warp and filling directions), which ultimately enables the separation and identification of the captured images. A computer program can rapidly process the data attained and determine the number and intensity of the fabric defects.