A non-expert organised visual database: a case study in using the Amazon metric to search images

Abstract

In a previous paper the notion of "using the Amazon metric to construct an image database based on what people do, not what they say" was introduced (see [1]). In that paper we described a case study setting where 20 participants were asked to arrange a collection of 60 images from most to least similar. We found they organised them in many different ways for many different reasons. Using Wexelblat's [2] semantic dimensions as axes for visualisation in conjunction with the Amazon metric we were able to identify common clusters of images according to expert and non-expert orderings. This second study describes the construction of a visual database based on the results of the first case study's non-expert participants' organising strategies and rationales. The same participants from the first study were invited to search for "remembered' images in the visual database. A better understanding was gained of their detailed reasonings behind their choices. This led to the development of a non-expert organised visual database that proved to be useful to the non-expert user. This paper concludes with some recommendations for future research into developing a non-expert, selforganising, visual, image database using multiple thesauri, based on these core studies.