A Study of Hierarchical Concatenation Networks in the Area of Pattern Recognition

Abstract

Hierarchical Concatenation Networks (HCN) are inspired by the way humans recognize patterns; i.e. by concatenating small features. In HCNs patterns are split into small parts, and then concatenated and activated in the network’s layers. The research in this thesis investigated and explored feature extraction methods, similarity measures, and classification using HCNs. Results indicate that HCNs can be used in automatic pattern recognition systems with better performance rate on the lower layer than the top layer.