A new model-based approach for industrial visual inspection

摘要

Two critical steps in industrial visual inspection are to design a flexible representation scheme and to match it with an effective inspection algorithm. Good representation and inspection algorithms can drastically speed up the subsequent matching process, and strengthen the feasibility of complicated on-line inspection. A novel pictorial representation called pattern attributed hypergraph (PAHG) and a new structural inspection algorithm are presented. The PAHG is used to describe the pictorial information of an input pattern, which includes attributed nodes, connectivity between nodes (intra-relations), regional primitives, and neighborhood relations between regions (inter-relations). The new structural inspection algorithm based on PAHG is proposed to match the inspected scene with the model efficiently. Consequently, the amount of matching computation is significantly reduced to roughly 1/k3 of the currently available method (Darwish and Jain, IEEE Trans. Pattern Analysis Mach. Intell.10, 56–68 (1988)), where k is the number of partitioned regions. The theoretical analysis and experimental results have shown the superiority of the proposed method.