Figure 1.
Two out of the 12 stimulus sets used in the experiment. Each row shows an ambiguous image in the first column with the two unambiguous versions next to it.
Figure 2.
Viewing behavior prior to awareness.
(a) Examples of viewing behavior prior to object awareness on the ambiguous and unambiguous stimuli with corresponding percepts. There are significant differences between the groups with different percepts (as indicated by the KL divergence analysis), and the differences in the viewing behavior on the ambiguous and unambiguous stimuli are aligned with identical percepts (as shown by the similarity index δ). The shown fixation patterns correspond to the fifth-largest index value out of the ten examined stimulus sets. (b) The cosine values underlying the similarity index calculation for the individual fixation density maps (FDM). Filled symbols represent percept A, the empty ones percept B. Squares denote cosines calculated from the unambiguous FDMs; diamonds indicate values calculated from the ambiguous FDMs. Image 1 corresponds to the example shown in (a). (c) The time-analysis showing the index-peak at about 1330 ms before the button press. Error bars are s.e.m. The shaded area marks the time during which data would be discarded according to the pupil analysis.
Figure 3.
Shown are the distributions of p-values for (a) KL-Divergence on the unambiguous stimuli, (b) KL-Divergence on Ambiguous Stimuli with different percepts, (c) the similarity index δ. All of them are nonuniform and right-skewed.
Figure 4.
Mean Support Vector Machine prediction accuracy for the correct percepts is shown for the ten tested image sets. Accuracy over fixations of one subject was calculated using SVM's trained on the remaining fixation data of the ambiguous stimuli (leave-one-out cross validation). Errorbars depict SEM's.