A Distributed Image-Database Architecture for Efficient Insertion and Retrieval Tasks

Title A Distributed Image-Database Architecture for Efficient Insertion and Retrieval Tasks
Author(s) R. Weber, H.-J. Schek
Type Article
In Fifth International Workshop on Multimedia Information Systems (MIS'99), Indian Wells, Palm Springs Desert, California, USA
Organization Institute for Information Systems, ETH Zurich
Month October
Year 1999


Allowing for content-based image retrieval, an image database must perform a number of time-consuming feature extraction tasks when inserting a new image, when searching with a newly entered image, or when bulk-loading many images in parallel. The duration of this preprocessing step typically lies far beyond what one would accept for conventional (short-running) database transactions. Hence, a natural step is to export the extraction tasks from the database and to distribute them among a number of external components in a cluster. As a consequence the database plays the role of a coordinator that assigns tasks to "its" auxiliary components. Since the execution times for feature extraction and the availabilities of such components vary over time, the coordinator needs additional information about the components in order to "optimally" assign extraction tasks to available components such that the overall preprocessing cost is minimized. At the same time, similarity search should be supported with high priority. In this paper, we present a coordination middleware (CoMid) on top of a distributed, component-based architecture that minimizes the cost of the preprocessing step and optimally assigns extraction tasks. We provide a solution for a dynamic environment: Any time a new feature extraction component can be added and any time a component can be switched off. Insertions as well as searches will not need to be restarted. The coordination middleware takes care of such "failures" and initiates "forward recovery". We describe in detail what data on the components and what image meta-data are necessary for the coordination, and how to exploit them. We report on an evaluation of this architecture using our prototype system applied to a very large image collection.

You can directly download a postscript (395 KB) version of this paper.
!!! Dieses Dokument stammt aus dem ETH Web-Archiv und wird nicht mehr gepflegt !!!
!!! This document is stored in the ETH Web archive and is no longer maintained !!!