Neural Pattern Recognition System Invariant to Rotation of Input Pattern and Its Application to Coin Recognition

Abstract

We have already presented a pattern recognition system insensitive to translation and rotation by 90° using neural networks. In this paper, we propose a pattern recognition system invariant to rotation of input pattern by any number of degrees. The system consists of a fixed invariance net with many slabs and a trainable multi-layered network. To illustrate the effectiveness of the system, we apply it to a variable rotation coin recognition problem between 500 yen and 500 won coins.