A new neural model for invariant pattern recognition

For most of the pattern recognition applications, it is often required to correctly recognize patterns even if they have variations in position, rotation, and/or scale. In this paper, to achieve the goal of invariant pattern recognition we propose a new neural model which consists of a cascade connection of four two-dimensional layers. The first three layers of the neural model perform the processes of position normalization, rotation normalization and feature extraction, respectively. The last layer is responsible for both recognition job and scale normalization by specially designing its output neurons to possess a scale invariant property. Finally, simulation results are given to demonstrate that the proposed model is simple and effective for invariant pattern recognition. Copyright (C) 1996 Elsevier Science Ltd