AUTOMATIC VISUAL INSPECTION OF WOVEN TEXTILES USING A 2-STAGE DEFECT DETECTOR

Automatic inspection of woven textile fabric is discussed. A two-stage detection process is adopted, with the second stage involving set of novel contextual decision fusion techniques. Three significant problem s are addressed: (1) texture feature extraction: Fourier transform fea tures are found to be well matched to the spatially periodic nature of the woven pattern; (2) detection of localized flaw patterns: since pr ior probabilities are impossible to estimate, and we cannot hope to en umerate all defect classes, a Neyman-Pearson approach is adopted, i.e. , flaw detection is via measured deviation from nominal; and (3) detec tion of extended flaw patterns: the most common flaws are characterize d by linear or other cluster shaped patterns; although these are weakl y detectable by local detectors, they may be ignored when local detect or sensitivity is set to achieve tolerably low false-alarm rates; a lo cal-extended contextual decision fusion technique using morphological filtering enables us to achieve very low composite false-alarm rate. T he performance of the system is evaluated on samples of denim fabric c ontaining real defects. The predicted composite false-alarm rate is of the order 1 in 10(13), Or equivalent to 1 per 100 km of fabric roll. Experimental results demonstrate the compatibility of this favorable f alse-alarm rate with the reliable detection of flaws, which have been chosen for their subtlety and detection difficulty. (C) 1998 Society o f Photo-Optical instrumentation Engineers.