A page from the OpenDepot.org service

Jump to the start of the main contents

Clarke, Alasdair D. F. and Dong, Xinghui and Chantler, Mike J. (2012) Does Free-sorting Provide a Good Estimate of Visual Similarity. In: Predicting Perceptions: Proceedings of the 3rd International Conference on Appearance. Lulu Press, Edinburgh UK, pp. 17-20. ISBN 978-1-4716-6869-2

PDF (Does Free-sorting Provide a Good Estimate of Visual Similarity) - Published Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (839Kb) | Preview


    The majority of work on texture analysis in computer vision has concerned texture classification and segmentation, while the problem of measuring and modelling the visual similarity between pairs of textures has been relatively neglected. One likely reason for this is the difficulty in collecting subjective human similarity judgments over a large database of textures. A common approach is to carry out a free-sorting experiment to obtain a similarity matrix which can then be mapped onto a low dimensional space using techniques such as MDS or Isomap. This results in a Euclidean space in which textures are represented as points, and the distance between two points is taken to represent the perceptual visual dissimilarity between the associated pair of textures. However, it is unknown if such a metric can generalise to predict human texture judgements in other tasks, or even if similarity judgements are metric at all. In this study we investigate this question by carrying out an experiment using a pair-of-pairs paradigm and compare these results to the predictions made by a low dimensional model (d = 3) obtained from a free-sorting experiment and find that it agrees with the judgements made by participants.

    Item Type: Book Section
    Subjects: UNSPECIFIED
    Divisions: UNSPECIFIED
    Depositing User: Dr Stefano Padilla
    Date Deposited: 04 May 2012 16:26
    Last Modified: 08 May 2012 12:20
    URI: http://opendepot.org/id/eprint/1043

    Actions (login required)

    View Item