Purchase this article with an account.
William Hahn, Elan Barenholtz; Alpha-Stable Distributions and Saccadic Foraging. Journal of Vision 2014;14(10):752. doi: https://doi.org/10.1167/14.10.752.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Given the limited perceptive range and informational capacity of the visual system, what is an optimal eye-movement strategy? While previous research has considered which image locations are informationally rich, much less research has considered general factors of saccadic movements, such as how often and how far the eyes should move under an optimal information-gathering strategy. With a higher resolution in the fovea and limited processing resources, the visual system must decide when an image region has undergone sufficient processing to move to a new location. Gaze shifts can thus be thought of as the visual system foraging for areas rich in visual saliency. Here we use statistical models to explore the overlap between simple animal foraging and gaze relocation. Research gathering behaviors of other species show that foragers try to minimize the distance travelled between targets to maximize their energy gain. An optimal foraging strategy must balance intensification with diversification, searching around the current solutions while making sure to explore the space efficiently. Heavy-tailed (alpha-stable) distributions have been shown to be advantageous when searching for randomly and sparsely distributed resources, yet research remains limited because closed form analytic expressions for these non-Gaussian distributions are not available. In the current study, we employ recently developed numerical methods to fit alpha-stable parameters to saccadic distributions. We find that the measured alpha values are predictive of scene geometry and distractor distribution, with measured alpha parameters falling in between Gaussian and Levy and decreasing as search difficulty increases. This suggests that eye movement distributions do not have closed form solutions and should be characterized by numerical approximations of stable distributions. Furthermore, these results demonstrate the potential utility of a new set of analytic tools for exploring this potentially rich source of behavioral data.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only