On Approximate Nearest Neighbour Field Algorithms in Template Matching for Surface Quality Inspection

Surface quality inspection is applied in the process of manufacturing products where the appearance is crucial for the product quality and customer acceptance, like for woven fabrics. The predominating approaches to detect defects are feature-based. Recently we investigated an alternative approach utilizing template matching in the context of regular or near-regular textured surface inspection. This paper reveals that the template matching approach belongs to the class of approximate nearest neighbour field (ANNF) algorithms which are common in a different field of image processing, namely structural image editing. By modifying a state-of-the-art ANNF algorithm the advantage of template matching algorithms for defect detection can be shown. Furthermore the importance of the chosen distance function is demonstrated in an explorative study and a concept to determine if the template matching approach is suitable for a given texture and defect type is demonstrated on a set of defect classes and texture types.