Defect Detection of Patterned Objects
Mechatronic systems are used in the production and inspection of fabrics. Monitoring these systems and detecting possible defects in the products are important in their proper operation. This chapter presents a comparison of the defect detection methods recently developed for a popular kind of patterned objects Jacquard fabric. These methods include golden image subtraction (GIS), wavelet-based defect detection, and Bollinger Bands (BB). The GIS method makes use of a golden image, which is larger than a repetitive unit, to perform a convolution filter on the fabric image based on the golden image. Not all types of defects can be successfully detected by the basic GIS method. Hence, a wavelet-preprocessed golden image subtraction (WGIS) method has been developed. The detection success rate of this method is as high as 96.7%, based on 60 images. A wavelet-based method, called Direct Thresholding on Detailed Subimages (DT), has been developed as well. The DT method has a detection success rate of 88.3% on 60 images. However, the detection method based on the use of BB is simple and very effective. In an extensive evaluation of 230 images, the detection success rate for this method is an impressive 99.57%.