Patterned Texture Inspection by the Regular Bands Method

Abstract

This paper presents a new defect detection method for patterned texture by the Regular Bands (RB) method. Patterned texture, such as patterned fabric, consists of a fundamental repetitive unit. The repetitive unit provides regularity features for every individual patterned texture and characterizes the design of the Regular Bands. A break in regularity is considered to be a defect in the Regular Bands method. The measures of moving average and standard deviation of the pixel intensity are used for defect detection. Defects, such as Broken End and Think Bar, have differential intensity changes compared to the pattern on a repetitive unit of the patterned texture. The Regular Bands method is an effective, fast and shift-invariant approach which can provide clear and clean defect segmentation on patterned fabric. Lots of previous defect detection methods focus on unpatterned (simple) fabric like plain and twill fabrics, while the Regular Bands method contributes to the patterned fabric. The proposed method has been evaluated on box-patterned fabric, with 98.21% detection success rate in a total of 56 defect-free and defective images.