A Dispatcher-driven Processing Architecture for Image Similarity Retrieval using Clustered Relaxation Matching servers

Abstract

A scalable dispatcher-driven processing architecture for image similarity retrieval that makes use of clustered relaxation matching servers is introduced. Literature reports on adapting the highly effective but computation intensive relaxation matching algorithm to on-line applications had primarily focused on architectures for parallel hardware implementation. In contrast, our architecture is software-based and achieves its scalability by using a tuneable dispatcher with variable number of relaxation matching servers executing concurrently on network connected workstations. Experimental results illustrating the scalability of the proposed architecture in terms of response time are given. The proposed architecture is not limited to using relaxation matching based servers, but is also applicable to other types of image matching server as well.