Multidimensional Scaling of Video Surrogates

This article is concerned with the problem of representing moving images for information retrieval. Of primary concern is evaluating the representativeness of different types of surrogates for various tasks. The basic factor considered is the ability of a surrogate to enable users to make the same distinctions that they would make given the actual video. To explore this issue, four types of video surrogates were created and compared under two tasks. Multidimensional scaling (MDS) was used to map the dimensional dispersions of usersʹ judgments of similarity between videos and similarity between surrogates. Congruence between these maps was used to evaluate the representativeness of each type of surrogate. Congruence was greater for image-based surrogates than for text-based surrogates overall, while congruence between text-based surrogates and videos was greatest when a specific task was introduced.