Pattern recognition using fragmentation and concatenation

This paper presents a pattern-recognition algorithm based on fragmentation and concatenation. The main mechanism was loosely inspired by Jeff Hawkins' Hierarchical Temporal Memory (HTM) model of the human neo-cortex [1]. In this algorithm pictures are fragmented and the fragments are statistically analysed and concatenated by a multi-layer network to obtain a kind of signature for the pattern. When new pictures are presented at the network's inputs their signatures are compared with previously stored ones to determine similarity. The network has been tested using a set of pictures of decimal digits. Preliminary results show that, in most cases, the network's ability to recognise patterns increases with the number of layers.