Artificial Intelligence and Query Execution Methods in the VizIR Framework

Abstract

The article introduces the architecture of the querying components of the visual information retrieval framework VizIR. A major design goal was to assure adaptability and extensibility in manifold ways. VizIR components can be arbitrarily combined to build extensive applications. The framework provides various visual content descriptors, similarity measures and query models. Moreover, the platform can be extended by visual features or similarity measures as well as entirely novel query paradigms. VizIR introduces an approach for Video Browsing based on MPEG-7 visual features and Self- Organizing Maps for clustering. Furthermore, a recently proposed approach for integrating browsing and retrieval techniques is presented. Following this approach, the user is enabled to interact with the querying system in several ways which improves retrieval quality and performance considerably.