TY - JOUR
T1 - The “where” and “when” of overt attention in natural scenes: fixation probability predicts fixation duration
AU - Einhäuser, Wolfgang
AU - Nuthmann, Antje
N1 - No DOI
PY - 2014
Y1 - 2014
N2 - The allocation of overt attention to a given location in a natural scene is characterized by two variables: the probability of the location to be fixated and the fixation's duration. While both measures have been used widely, surprisingly little is known about their relation. Using a large corpus of fixation data of 72 participants, 135 colour images and 3 distinct tasks, we first demonstrate that fixation probability ("empirical salience") computed on a subset of participants predicts fixation duration in a disjoint subset performing a different task. Linear-mixed-effect modelling shows that this relation cannot be explained away by joint dependencies on low-level features (e.g., luminance, contrast, edge density) nor by spatial biases. In a second step, we demonstrate that fixation probability obtained on these corpus data still predicts fixation durations for novel experimental data, even if stimuli are deprived of colour, globally reduced in luminance or contrast, or locally equalized in either luminance or contrast. Only if object content is removed, while keeping second-order statistics intact, the relation between fixation probability and duration falters. In sum, our data show a robust dependence of fixation duration on fixation probability, which does not depend on low-level image features, but on high-level scene content.
AB - The allocation of overt attention to a given location in a natural scene is characterized by two variables: the probability of the location to be fixated and the fixation's duration. While both measures have been used widely, surprisingly little is known about their relation. Using a large corpus of fixation data of 72 participants, 135 colour images and 3 distinct tasks, we first demonstrate that fixation probability ("empirical salience") computed on a subset of participants predicts fixation duration in a disjoint subset performing a different task. Linear-mixed-effect modelling shows that this relation cannot be explained away by joint dependencies on low-level features (e.g., luminance, contrast, edge density) nor by spatial biases. In a second step, we demonstrate that fixation probability obtained on these corpus data still predicts fixation durations for novel experimental data, even if stimuli are deprived of colour, globally reduced in luminance or contrast, or locally equalized in either luminance or contrast. Only if object content is removed, while keeping second-order statistics intact, the relation between fixation probability and duration falters. In sum, our data show a robust dependence of fixation duration on fixation probability, which does not depend on low-level image features, but on high-level scene content.
M3 - Meeting abstract
SN - 0301-0066
VL - 43
SP - 11
EP - 11
JO - Perception
JF - Perception
T2 - 37th European Conference on Visual Perception
Y2 - 24 August 2014 through 28 August 2014
ER -