Bridging the semantic gap in visual information retrieval
This project was conducted between 2004 and 2007 by a team drawn from the universities of Brighton and Southampton. It sought to bring new understandings and competencies to the problem of retrieving still images from within large, managed collections of such artefacts. The existence of a ‘semantic gap’ is a well-known limitation on the functionality of present-day visual image retrieval systems. Given the vast and constantly expanding scale upon which visual resources are made available via the World Wide Web, users and providers of images, as well as the research communities, have much to gain from the bridging of this semantic gap. The project aimed to develop a fully informed view of the semantic gap and the functionality to be gained by visual image retrieval systems in bridging it. This aim was addressed by using a broad spectrum of operational image retrieval activity as a survey base, and analysing the semantic content of images articulated by an advanced, faceted conceptual model of semantic content. The potential which cross-paradigm metadata offers for bridging the semantic gap was explored in experimental systems, a particular realization of which in the form of a novel semantic-space model was shown to have encouraging spanning potential.
Notwithstanding the power of the experimental system developed in the course of the project, the work provided fresh insights into the true scale of the semantic gap in image retrieval, the further reaches of which remain beyond the current capability of systems which seek to replace human vision by computer vision.