Model based defect detection problem: Particle filter approach

Abstract

This paper addresses the raw textile defect detection problem. An efficient algorithm based on Bayesian estimation is presented for detection of defects encountered in textile images. Bayesian estimation is performed by particle filtering. Performance inprovement in detection rate has been verified through extensive computer simulations.

Patch based yarn defect detection using Gabor filters

“…At the same time, for R1 fabrics, the image and patch misdetection rate becomes negligible. Globally, the detection rate increases up to 98.8% from 97% of Basybuyuk et al [15]. A decrease in performance is found for E2 defect class in R3 yarn, because this type of defects (dirt spots) is particularly critical in highly structured yarns, as confirmed by the results obtained for the human inspectors.…”

“…On the other side, the use of alg1 b causes a performance degradation of about 10%. Comparisons between the algorithms presented in [14], [15], and this work, considering the TILDA database, have been reported in Table I. In this table, P fa with respect to E0, and P md,I with respect to E1, E2, E3, and E4 defect classes, for R1 and R3 fabric types, have been considered and compared with the values reported in [14].…”

“…In this table, P fa with respect to E0, and P md,I with respect to E1, E2, E3, and E4 defect classes, for R1 and R3 fabric types, have been considered and compared with the values reported in [14]. The comparison with [15] has been obtained defining a detection rate parameter as the number of patches correctly classified as nondefective added to the number of patches correctly classified as defective, divided by the total number of tested patches.…”

“…A decrease in performance is found for E2 defect class in R3 yarn, because this type of defects (dirt spots) is particularly critical in highly structured yarns, as confirmed by the results obtained for the human inspectors. [14], [15], AND THE PROPOSED ALGORITHM ("-" MEANS THE INFORMATION IS NOT AVAILABLE).…”

Automated defect detection in uniform and structured fabrics using Gabor filters and PCA

“…Comparisons between the algorithms presented in [34], [35], and the algorithms alg and alg K=4S=4 , considering the TILDA database, are reported in Table 1. P fa with respect to E0, and P md,I with respect to E1, E2, E3, and E4 defect classes, for R1 and R3 fabric types, are considered and compared with the values reported in [34].…”