An approach to distributed model selection for multimedia content-based indexing in a decentralized context

A major need of multimedia indexing and retrieval on a P2P network is related to characterisation of classes. Currently, many multimedia retrieval systems have been proposed and developed but these systems are not designed to be distributed across different computers on a network. We focus on the case that many retrieval systems exist on the network where all densities are Gaussian mixtures due to its numerous good properties. In this paper, we have proposed an algorithm which uses a combination of cross validation and merging to find the best model among many models which exist on the network for a unique class .