The latest issue of Journal of Documentation includes an interesting paper on content-based image retrieval. This describes a survey of operational image retrieval activity and a sample collection of user requests, the resulting images retrieved and their associated metadata. The research was undertaken as part of the ‘Bridging the semantic gap in visual information retrieval’ project funded by the Arts & Humanities Research Council. The purpose of the research is stated as:
“To provide a better-informed view of the extent of the semantic gap in image retrieval, and the limited potential for bridging it offered by current image semantic retrieval techniques.”
Enser, Peter G. B. & Sandom, Christine J. (School of Computing, Mathematical and Information Sciences, University of Brighton; Hare, Jonathon S. & Lewis, Paul; H. (School of Electronics and Computer Science, University of Southampton). Facing the reality of semantic image retrieval. JDoc 63(4), 2007, p465-481.
Excerpt from the Introduction
“For those engaged in the commercial exploitation of visual image material, or whose curatorial responsibilities address our visual heritage, the concept of ‘semantic image retrieval’ is one with which they have a long-standing familiarity, albeit that the term is not within their normal working vocabulary. For those within the computer science and computer vision communities, however, the term ‘semantic image retrieval’ represents the frontiers of current thinking and practice.”
“This paradox reflects the unfortunate fact that the endeavours of the latter communities have been little informed by the needs of real users or the logistics of managing large scale image collections; correspondingly, among the practitioner community, there has been only a minimal engagement with the endeavours of those engaged in image retrieval research.”