Neural network image segmentation for automated visual inspection

A fundamental problem, not satisfactory solved for automated visual inspection, is the segmentaiton of images. To overcome the segmentation problem we suggest a fast and intelligent segmentation for advanced vision systems. The main idea of our approach is to combine the power of textgure segmentaiotn with the ability to learn of neural networks. In this paper we focus on the evaluation of different neural netowrk model.s Two kinds of feed-forward networks are compared: The multi-layered perceptron MLP and the restricted coloumb energy model (RCE). To evaluate the performance for industrial applications we calculate not only the rate of correct/incorrect classifications but we also take into consideration the computatiuonal effort and the possibility of retraining and rejection.