PUB - Publikationen an der Universität Bielefeld

A computer vision system which can be trained to classification tasks from sample views is presented. It consists of several artificial neural networks which realize local PCA with subsequent expert nets as classifiers. The major benefit of the approach-is that entirely different tasks can be solved with one and the same system without modifications or extensive parameter tuning. Therefore, the architecture is an example for the potential which lies in view based recognition: Making complicated tasks solvable with less and less expert knowledge.