On-chip template training for pattern matching by cellular neural network universal machines (CNN-UM)

Pattern matching problems using statistical methods generally result in high computational effort. On the other side algorithms based on CNN technology can provide efficient new solutions for complex image processing tasks. In various applications template values are determined by an optimization procedure using simulation systems. In this contribution an optimization method directly interacting with a CNN-UM chip will be presented to treat a CNN based pattern matching problem. Thereby a certain binary pattern of an image also comprising other different patterns should be extracted. The proposed on-chip training leads to highly adapted templates solving the given tasks in different setups.