In this paper, we propose a framework for the usage of a multi-agent architecture to perform automatic indexing of multimedia documents. Because of the diversity of the data types that are involved in multimedia documents, multimedia indexing requires the combination of very different technical approaches: rule-based, syntax-based, statistical, probabilistic, computational etc... We propose a framework where we can combine simple agents (generally based on a single media) to build a system which is able to perform the automatic detection of complex multimedia events. We address three issues in this paper. First, we give a general overview of the organization of a multi-agent system, and define the various ways by which simple agents can be combined. Then we propose a probabilistic formulation that allows to combine the results of the computation of these agents, and in particular, is able to handle the uncertainties that arise in the recognition processes. Finally, we give some indications on how distributed search algorithms could be used to efficiently handle the computations that are involved by these agents. Throughout the paper, we will consider a typical example of the kind of indexing that we would like to be able to achieve, to serve as an illustration for the various aspects that we are proposing.
A framework for multi-agent multimedia indexing
IJCAI workshop on intelligent multimedia information indexing, August 1995, Montreal, Canada
PERMALINK : https://www.eurecom.fr/publication/141