A Parallel, Distributed and Associative Approach for Searching Image Patterns with Holographic Dynamics

Abstract

This paper presents a new associative pattern-matching network based on a digital adaptation of optical holography. Unlike any existing neural network or associative memories, it can localize its search dynamically on any subset of the pattern space and at the same time generate a feedback on the quality of the match. Current associative memories based on neuro computing are unable to support such meta-interactions. The scheme involves adaptive ‘enfolding’ of the raw massive search space into a holograph and direct regeneration of the matched target pattern during a search. The search process is a constant-time operation compared to traditional algorithm approaches, inherently parallelizable, and is an excellent candidate for hardware or optical implementation. This new technique is expected to facilitate significantly applications that require direct pattern matching in massive image repositories in real time.Target recognition,visual query,content-based image retrieveandautomatic index-extractionare just a few of such applications.