Defect detection on the fabric with complex texture via dual-scale over-complete dictionary

This paper proposed a defect detection algorithm for fabrics with complex texture based on dual-scale over-complete dictionary. The core was to learn the features of defect-free fabrics using dual-scale over-complete dictionary. The traditional defect detection methods generally showed a favorable effect on plain cloth or twill, etc., while poor effects on plaids or stripes, etc. with complex texture. Considering the large variations of the defect size of different kinds, this study used dual-scale dictionary to enhance self-adaptability of defect detection. Subsequently, the increase in the false detection rate brought by dual-scale detection was effectively avoided using fusion algorithm of different scales. The experiment based on TILDA database suggested that the algorithm proposed achieved a detection rate of 96.5% and a false detection rate of 5.5% for complex texture. Moreover, this algorithm showed favorable self-adapting ability to other fabrics on our own database. Through the downsampling operation on large-scale samples, the computing time was greatly reduced as compared to single-scale algorithm.