Abstract
In this paper, a three-layered pattern-rotation model is proposed, which automatically generates rotated patterns from an input pattern. This model has the capability of rotating input patterns both at the stage of learning and recognizing, and has the merit that the loads of learning and recognizing can be freely adjusted. As for this model, if more rotated patterns are generated at the stage of learning, the load of learning becomes high but the load of recognizing decreases. Inversely, if less rotated patterns are generated at the stage of learning, the load of learning becomes low but the load of recognizing increases.
In the proposed model, it has following merits: its design is easier, the net size is smaller, learning and recognizing time is shorter than the conventional models that have a fixed rotation invariant net followed by an adaptive net.
