Efficiently Managing Multimedia Hierarchical Data with the WINDSURF Library

Abstract

Complex multimedia data are at the heart of several modern applications, such as image/video retrieval and the comparison of collection of documents. Frequently, such complex data are modeled as hierarchical objects that consist of different components, like videos including shots, images including visually coherent regions, and so on. When such complex objects are to be compared, for example, for assessing their mutual similarity, this is usually done by recursively comparing component elements. However, due to such complexity, it is often hard to efficiently perform a number of tasks, like processing of queries or understanding the impact of different alternatives available for the definition of similarity between objects. In this article, we propose a unified model for the representation of complex multimedia data, introducing the WINDSURF software library, with the goal of allowing a seamless management of such data. The library provides a framework for evaluating the performance of alternative query processing algorithms for efficient retrieval of multimedia data. Important features of the WINDSURF library are its generality, flexibility, and extensibility. These are guaranteed by the appropriate instantiation of the different templates included in the library: in this way, each user can realize her particular retrieval model of need.

This work was partially supported by the CoOPERARE MIUR Project.