A pattern recognition based Holographic Graph Neuron for Persian alphabet recognition

Abstract

In this article a Vector Symbolic Architectures is purposed to implement a hierarchical Graph Neuron for memorizing patterns of Persian/Arabic isolated characters. The main challenge in this topic is using Vector Symbolic representation as a one-layered design for neural network while maintaining the previously reported properties and performance characteristics of hierarchical Graph Neuron. The designed architecture is robust to noise and enables a linear (with respect to the number of stored entries) time search for an arbitrary sub-pattern. The proposed method was implemented on a standard Persian database and the obtained results showed the ability of (not necessarily better) Graph neuron to recognize the Persian isolated character patterns.