A new self-organizing neural model for invariant pattern recognition

摘要

This paper proposes a new neural model for invariant pattern recognition. The proposed model can efficiently recognize patterns regardless of their possible variations of position, rotation and scale. It consists of two systems: feature detection and classification. The feature detection system consists of three two-dimensional layers: positon normalizaton, rotation normalization and feature extraction layers. These three layers take the responsibility of position normalization, rotation normalization and feature extraction, respectively. Classification system is a two-layer feed-forward neural network and its function is to perform classification as well as scale normalizaton. The proposed model is shown to be effective for invariant pattern recognition. Finally, simulation results are given for demonstration.