[ Potential applications of this research: Quality control in textile, ceramics, tile, wallpaper, printed circuit board, aircraft window industries and latest 3D printing ]
There have been numerous research for defect detection on unpatterned texture (Fig. 1(a)), but rarely on patterned texture (Fig. 1(b)(c)). Through many in-depth studies of patterned textures (fabric) [3-6], I developed three generic defect detection methods [3-5] and then proposed a unified framework [6-8] for defect detection based on different features of patterned textures. Little research in patterned texture defect detection has been carried out before my research due to several reasons: (1) complicated patterns on patterned textures compared to unpatterned textures; (2) numerous categories of patterned textures; (3) defects having a high degree of similarity in shape with respect to the texture background.
The biggest contribution of my previous research on automated defect detection of patterned textures has three folds. First, I developed three effective, robust and promising methods: wavelet-based golden image subtraction (WGIS) method (96.7% accuracy) , Bollinger bands (BB) method (98.59% accuracy)  and Regular bands (RB) method (99.4% accuracy) . Second, these methods are a breakthrough in defect detection because they could explicitly outline and identify the geometrical shapes of various defects. Fig. 2(a) shows a sample of dot-patterned fabric with thin bar defect, in which the detection results of WGIS (Fig. 2(b)), BB (Fig. 2(c)), RB (Fig. 2(d)) are illustrated. Third, I developed a profound theoretical framework of generalized motif-based defect detection method  for 16 out of 17 wallpaper groups of all two-dimensional (2D) patterned textures (Fig. 3(a)). The motif-based method does not need a ground truth to be cross-referenced and this property is superior to most supervised defect detection methods. An energy-variance (EV) space (Fig. 3(b),(c)), generated by the inputted motifs, was constructed with Max-Min decision region (MMDR). This method reached 93.86% detection accuracy for 16 wallpaper groups in an extensive evaluation . A review paper, co-authored with my doctoral supervisors, in automated fabric defect detection can be found in the Journal of Image and Vision Computing  in 2011.
Regarding pattern recognition techniques, I modified MMDR into ellipsoidal decision region (EDR)  (Fig. 4) to improve ill detection of some false-positive (FP) and false-negative (FN) cases. The distributions of EV values were assumed to conform to Gaussian mixture model (GMM). With k-mean clustering, appropriate number of EDRs could be determined and more accurate detection accuracies were achieved in the p2 and pmm groups with respect to MMDR.
Recently, I have collaborated with the colleagues at HKBU to develop a novel fabric inspection and visualization by the method of image decomposition . In this decomposition, we assume a defective patterned fabric image (Fig. 5(a)) is a superposition of defective objects (cartoon structure in Fig. 5(b)) and patterned texture (texture structure in Fig. 5(c)). This new method is an early attempt to exploit an image decomposition model by a variable splitting method in convex optimization area and offers a satisfied accuracy of 95%-99% for most testing images.
More recently, a modified Bollinger Bands with directional iterations (Fig. 6)  has been developed for the plain and twill fabric defect detection. The detection accuracies for 77 defective images and 100 defect-free images are 96.1% and 96%, respectively. In a pixel-to-pixel evaluation comparing the detection results of the defective images with the ground-truth images, a 93.51% detection success rate is achieved. Also, a new inspection method by an Elo rating system  which applies the spirit of sportsmanship has been developed with my final year project’s student, Colin Tsang, in 2014. The fabric defect detection was achieved by fair matches between any two partitions on every detected image and an overall detection success rate of 97.07% was obtained.
 H.Y.T. Ngan, “Patterned Jacquard Fabric Defect Detection,” M.Phil. Thesis, The University of Hong Kong, 2004.
 H.Y.T. Ngan, “Motif-based Method for Patterned Texture Defect Detection,” Ph.D. Thesis, The University of Hong Kong, 2008.
 H.Y.T. Ngan, G.K.H. Pang, S.P. Yung, and M.K. Ng, “Wavelet based methods on Patterned Fabric Defect Detection,” Pattern Recognition, vol. 38, issue 4, pp. 559-576, 2005.
 H.Y.T. Ngan, and G.K.H. Pang, “Novel Method for Patterned Fabric Inspection using Bollinger Bands,” Opt. Eng., vol. 45, no. 8, 2006.
 H.Y.T. Ngan, and G.K.H. Pang, “Regularity Analysis for Patterned Texture Inspection,” IEEE Trans. Automation Science & Engineering, vol. 6, no. 1, pp. 131-144, 2009.
 H.Y.T. Ngan, G.K.H. Pang, N.H.C. Yung, “Motif-based Defect Detection for Patterned Fabric,” Pattern Recognition, 41(6), pp. 1878-1894, 2008.
 H.Y.T. Ngan, G.K.H. Pang and N.H.C. Yung, “Performance Evaluation for Motif-based Patterned Texture Defect Detection,” IEEE Trans. Automation Science & Engineering, vol. 7, no. 1, pp. 58-72, 2010.
 H.Y.T. Ngan, G.K.H. Pang and N.H.C. Yung, “Ellipsoidal Decision Regions for Motif-based Patterned Fabric Defect Detection,” Pattern Recognition, vol. 43, no. 6, pp. 2132-2144, 2010.
 H.Y.T. Ngan, G.K.H. Pang and N.H.C. Yung, “Automated Fabric Defect Detection – A Review,” Image & Vision Computing, vol. 29, no.7, pp. 442-458, 2011.
 M.K. Ng, H.Y.T. Ngan, X. Yuan and W. Zhang,”Patterned Fabric Inspection and Visualization by the Method of Image Decomposition,” IEEE Trans. Automation Science & Engineering, vol. 11, no. 3, pp. 943-947, 2014 .
 H.Y.T. Ngan and G.K.H. Pang, “Robust Defect Detection in Plain and Twill Fabric Using Directional Bollinger Bands,” Optical Engineering, vol. 54, no. 7, 073106, 2015. doi:10.1117/1.OE.54.7.073106.
 C.S.C. Tsang, H.Y.T. Ngan and G.K.H. Pang, “Fabric Inspection based on the ELO Rating Method,” Pattern Recognition (Accepted).