Scanpath similarity depends on how you look at it: Evaluating a ‘MultiMatch’ comparison algorithm

  • Richard Dewhurst (Speaker)
  • Marcus Nyström (Speaker)
  • Jarodzka, H. (Speaker)
  • Kenneth Holmqvist (Speaker)

    Activity: Talk or presentation typesOral presentationAcademic

    Description

    Visual scanpaths represent a paradox; they are incredibly revealing yet inherently difficult to compare. Here we evaluate our new method for comparing scanpaths (Jarodzka et al., 2010, Proceedings of the 2010 Symposium on Eye-Tracking Research & Applications, 211-218). Instead of representing fixation-and-saccade sequences by way of discreet Regions of Interest (ROIs) and comparing strings of corresponding letters for similarity (cf. Levenshtein string-edit distance), or representing scanpaths as Gaussian-based attention map functions which retain no information about the order of fixations, our ‘MultiMatch’ method treats scanpaths as geometric vectors and performs pair-wise comparisons based on a number of dimensions: shape, position, length, direction and duration. This requires a simplification step using thresholding to align scanpaths, and enables us to quantify characteristics which other comparison principles find problematic. With data from two experiments we assess how this algorithm copes with within- and between-subject similarity calculations, how task difficulty affects scanpath similarity, and how the similarity metric produced is dependent on scanpath length. Scanpaths can be similar in a number of intuitive ways, and the capability of our method to use multiple dimensions allows these qualitative distinctions to be tested empirically.
    PeriodAug 2011
    Event titleEuropean Conference on Eye Movements 2011
    Event typeConference
    Conference number16
    LocationMarseille, France