Real-time aspects of SOM-based visual surface inspection

摘要: We have developed a self-organizing map (SOM) -based approach for training and classification in visual surface inspection applications. The combines the advantages of non-supervised supervised offers an intuitive user interface. is less sensitive to human errors, since labeling large amounts individual samples not necessary. In classification, interface allows on-line control class boundaries. Earlier experiments show that our gives good results wood inspection. this paper, we evaluate its real time capability. When quite simple features are used, bottleneck nearest SOM code vector search during phase. experiments, compare acceleration techniques suitable high dimensional neighbor typical method. even can improve speed considerably, be used with standard PC.