Fabric defect detection using multi-level tuned-matched gabor filters

Kai Ling Mak, Pai Peng, Ka Fai Cedric Yiu

Research output: Journal article publicationJournal articleAcademic researchpeer-review

27 Citations (Scopus)


This paper proposes a new defect detection scheme for woven fab- rics. The proposed scheme is divided into two parts, namely the training part and the defect detection part. In the training part, a non-defective fabric image is used as a template image, and a finite set of multi-level Gabor wavelets are tuned to match the texture information of the image. In the defect detection part, filtered images from different levels are fused together and the constructed detection scheme is used to detect defects in fabric sample images with the same texture background as that of the template image. A filter selection method is also developed to select optimal filters to facilitate defect detection. The nov- elty of the method comes from the observation that a Gabor filter with finer resolutions than the fabric defects yarn can contribute very little for defect segmentation but need additional computational time. The proposed scheme is tested by using 78 homogeneous textile fabric images. The results exhibit ac- curate defect detections with lower false alarms than using the standard Gabor wavelets. Analysis of the computational complexity of the proposed detection scheme is derived, which shows that the scheme can be implemented in real time easily.
Original languageEnglish
Pages (from-to)325-341
Number of pages17
JournalJournal of Industrial and Management Optimization
Issue number2
Publication statusPublished - 1 May 2012


  • Defect detection
  • Industrial inspection
  • Multi-level Gabor wavelet
  • Woven fabrics

ASJC Scopus subject areas

  • Business and International Management
  • Strategy and Management
  • Control and Optimization
  • Applied Mathematics


Dive into the research topics of 'Fabric defect detection using multi-level tuned-matched gabor filters'. Together they form a unique fingerprint.

Cite this