Recognition of Hand Written Patterns by Rotation-Invariant Neocognitron.
Shunji Satoh, Jousuke Kuroiwa, Hirotomo Aso, Shogo Miyake · 1998
A rotation-invariant neocognitron, which has been recently proposed by authors, is trained by using hand-written numerical patterns provided by ETL-1 database. A new learning algorithm, “auto-generating algorithm”, is proposed and the learning time is reduced. The model can recognize realistic hand-written patterns. It is also shown that the model is completely robust for rotations of hand-written patterns. 1