Fully automated approach for patterned fabric defect detection

This paper introduces a fully automated approach for defect detection in patterned fabrics. First, the fabric pattern period is determined by calculating the image mean vectors in both horizontal and vertical directions. Second, Gray Level Co-occurrence matrices are calculated to both the reference defect-free image and defected image after dividing them to blocks that have the same dimension as the fabric pattern. Third, Euclidean distances are calculated between each gray level co-occurrence matrix and a reference one for both defect-free and defected images. Forth, the resultant Euclidean distances are compared to pre-calculated thresholds to identify the defected blocks. The experimental results show that the proposed algorithm can achieve high detection accuracy rate besides its simplicity.