A Data Allocation Method for Efficient Content-Based Retrieval in Parallel Multimedia Databases

摘要

Scaling up to large multimedia databases with high dimensional metadata descriptions while providing fast content-based retrieval (CBR) is getting increasingly important for many applications. To address this objective, we strive to exploit the popular parallel shared-nothing architecture. In this context, a major problem is data allocation on the different nodes in order to yield efficient parallel content-based retrieval. In this paper, assuming a clustering process and based on a complexity analysis of CBR, we propose a data allocation method with an optimal number of clusters and nodes. We validated our method through experiments with different high dimensional synthetic databases and implemented a query processing algorithm for full k nearest neighbors.