A novel self-organizing neural network for defect image classification

In this paper a novel self-organizing neural network called the Evolving Tree is applied to classification of defect images. The Evolving Tree resembels the Self-Organizing Map (SOM) but it has several advantages over the SOM. Experiments present a comparison between a normal SOM, a supervised SOM, and the Evolving Tree algorithm for classification of defect images that are taken from a real web inspection system. The MPEG-7 standard feature descriptors are applied. The results show that the Evolving Tree provides better classification accuracies and reduced computational costs over the normal SOMs.