Machine learning in hypermedia using digital image content

Radhakrishnan, Periasamy (1999) Machine learning in hypermedia using digital image content. University of Southampton, Doctoral Thesis.

Abstract

Visual documents are a powerful medium for the human cognitive process. The explosive proliferation of digital imagery and the development of information space in the form of the World Wide Web are forcing us to consider automated tools to manage visual documents similar to text documents.

This thesis investigates the use of agent-like processes with machine learning and classification capabilities to enhance content- and concept-based retrieval and navigation for multimedia information. A modified architecture for enhancing multimedia information handling is proposed and the use of agent-like processes within the architecture is described.

The thesis explores a variety of image representation techniques including invariant moments, Zernike moments, the wavelet transforms and novel combinations of these representations. Their merits as feature vectors in image classification are assessed.

Machine learning techniques, including Learning Vector Quantisation (LVQ) and Artificial Neural Networks (ANNs), are evaluated as candidates for image classification applications in the agents. An improved variant of Decision Surface Mapping (DSM) is proposed and its superior performance demonstrated over ANNs trained with error back-propagation for classifying grey-level images.

The thesis concludes that media content- and concept-based hypermedia systems with agent-based facilities show promise for meeting some of the challenge in content-based information retrieval and navigation.

This record has no associated files available for download.

Contributors

Download statistics