December 2022
Volume 22, Issue 14
Open Access
Vision Sciences Society Annual Meeting Abstract  |   December 2022
Using observer similarity matrices to understand individual differences in gaze behaviour towards objects in complex scenes
Author Affiliations & Notes
  • Marcel Linka
    Justus-Liebig-Universität Gießen
  • Benjamin de Haas
    Justus-Liebig-Universität Gießen
  • Footnotes
    Acknowledgements  This research was supported by European Research Council Starting Grant 852885 INDIVISUAL; BdH was further supported by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) Project No. 222641018–SFB/TRR 135 TP C9
Journal of Vision December 2022, Vol.22, 3424. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Marcel Linka, Benjamin de Haas; Using observer similarity matrices to understand individual differences in gaze behaviour towards objects in complex scenes. Journal of Vision 2022;22(14):3424.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Previous research revealed stable individual gaze biases along several semantic dimensions when viewing complex scenes. For instance, some observers have a stronger tendency to fixate faces while others fixate text more often. However, it is unclear what proportion of individual gaze differences towards objects in complex scenes is explained by these known dimensions. Here, we used a novel approach to analyse individual differences in gaze behaviour by computing ‘dimension-agnostic’ (dis-)similarities between observers. We analysed an eyetracking data set of N = 103 subjects, who freely viewed 700 complex scenes, depicting > 5000 objects. To estimate the noise ceiling of stable individual differences independently of a priori dimensions, we computed the similarity (Euclidean distance) of dwell times across all objects for each pair of observers. These pairwise similarities were highly consistent across independent subsets of images (R2 = .81). To explore how much of these stable differences can be explained by known individual gaze tendencies towards semantic dimensions, we determined individual fixation tendencies towards objects of 5 semantic dimensions and computed the similarity between each pair of observers for each dimension. We then fitted a cross-validated ridge regression model, predicting overall observer similarity from similarity matrices for each dimension. Results indicated that known dimensions captured 27% of the explainable variance in object fixations. Individual tendencies to fixate Faces (β = .22) and Text (β= .17) carried the highest weights, followed by fixation biases for Taste (β= .11), Motion (β= .10) and Touched (β= .07). Together, our findings suggest that individual tendencies along known semantic dimensions account for a substantial part of individual differences in object fixations. At the same time, most of the explainable variance in object fixations is not captured by these dimensions. We currently explore further dimensions, including individual biases for lower level image features and saccadic behaviour.


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.