Content-Based Trademark Recognition and Retrieval Based on Discrete Synergetic Neural Network

Abstract

Synergetic Neural Network (SNN) as proposed by Hermann Haken is a novel top-down, self-organized system. In this chapter, its associated discrete SNN is proposed and the recognition stability and the convergence of a generalized discrete SNN is analyzed. We proposed an adaptive algorithm of iterative step length refinement for synergetic recognition, which can ensure fast convergence and network steadily for all kinds of input pattern. Additionally, we apply the SNN to trademark retrieval and study its ability to support affine invariant retrieval of 2D patterns. To this end, we propose an affine invariant input vector in the frequency domain for the SNN and evaluate the retrieval ability of such networks for different types of input queries, for example, query by complete trademark pattern and query by image components. We show experimentally that our proposed SNN method is noise tolerant as well as able to support affine invariant retrieval. This led us to propose a novel paradigm for trademark retrieval based on visual keywords whereby trademark images can be queried in terms of simple geometric components.