Figure 1.
(A) An example trajectory recorded during the experiment. The fixation labeled with zero is the first fixation in that trial, which was excluded from analyses. (B) Distributions of distances between fixation locations and the closest bubble center for “first fixations” into a bubble (median 1.05°) and “subsequent fixations” within the same bubble (median 0.91°). For comparison, the distribution that would result if all fixations were sampled from the Gaussian window used to construct the bubbles (median 1.18°) is also given.
Figure 2.
Stimulus information versus number of bubbles for the four tasks.
Stimulus information estimated using the p-model is plotted for all four tasks (black dashed line). This is contrasted with the measured stimulus information in the same condition (green line) and in the congruent condition (red line). The blue line marks the measurements that result if the positions of bubble stimuli of the same condition are shuffled (same, permuted). The colored stars mark significant differences (p<0.05, bootstrapped confidence intervals) between the curve belonging to the respective condition and the p-model estimate. For visibility, the 95% confidence interval is marked by error bars only for condition same.
Figure 3.
Relationship between stimulus dependent and empirical salience.
(A) Example stimulus from the expression task with bubbles labeled by their stimulus dependent salience. (B) Scatter plot of stimulus dependent vs. empirical salience for the expression task. The positions of the bubbles from the example stimulus are marked by colored dots. The correlation coefficient r is given as a figure inset. (C) Correlation coefficients for all four tasks (E – expression, G – gender, I – influence, S – space). One star marks a significant correlation (p<0.05, t-test); two stars mark a highly significant correlation (p<0.01, t-test).
Figure 4.
(A) Example stimulus from the expression task where the individual bubbles are labeled by their fitted response distributions and the corresponding bubble information. The four numbers above the black line give the response probabilities for the classes “disgusted,” “happy,” “fearful,” and “sad.” The bold number below the line gives the bubble information (in bit). For the whole stimulus, the measured response distribution and stimulus information (in bit) is given in the lower right corner. (B) The distribution of bubble information for the expression task. The bubble information of the four bubbles of the example stimulus is marked by colored dots. (C) The distribution of bubble information for the other three tasks gender, influence and space.
Figure 5.
Relationship of task dependent and empirical salience.
(A) Example stimulus of the expression task with individual bubbles labeled by their bubble information. (B) Scatter plot of bubble information and empirical salience for the expression task. The positions of the example bubbles are marked by colored dots. The correlation coefficient r is given as a figure inset. (C) Correlation coefficients for all four tasks. Two stars mark highly significant correlations (p<0.01, t-test).
Figure 6.
Relationship of spatial bias and empirical salience.
(A) Example stimulus of the expression task with individual bubbles labeled by their spatial bias salience. (B) Scatter plot of spatial bias and empirical salience for the expression task. The positions of the example bubbles are marked by colored dots. The correlation coefficient r is given as a figure inset. (C) Correlation coefficients for all tasks. Two stars mark highly significant correlations (p<0.01, t-test).
Figure 7.
Relationship between empirical salience, stimulus dependent salience, bubble information and spatial bias salience for an example stimulus.
The example stimulus from the expression task is given on the left. The four values characterizing each bubble are shown on their respective scales (right panel). The range of spanned values for each variable is mapped to the same interval for comparison. The colors code for the identity of the different bubbles.
Figure 8.
Influence of the three factors on empirical salience.
The multivariate regression results are given for all four tasks expression (E), gender (G), influence (I), and space (S). The height of each bar depicts the R2 value; each shaded area represents the squared semi-partial correlation coefficient, which reflects the unique contribution of the respective factor. The white area in each bar represents the amount of variability of empirical salience that can explained by more than one factor.
Table 1.
Results of the multivariate regression.
Figure 9.
The different stimulus classes.
Subjects had to classify faces and forest scenes according to four tasks (expression, gender, influence, and space). For the forest scenes, the different response possibilities are given above the example stimuli. The stimuli are shown as full fields and are used for bubble stimuli construction. For copy right reasons, we cannot show the face stimuli here but we refer the reader to Tottenham et al. [78]. The face stimuli are taken from the “NimStim” stimulus set.
Figure 10.
(A) Distribution of bubble positions for the expression task. (B) A single bubble based on a patch of 6 visual degrees from a full field face stimulus. The patch was filtered using an eccentricity dependent frequency filter simulating the drop of spatial acuity and a Gaussian mask to avoid edge effects. (C) Different types of bubble stimuli were generated. Stimuli of the same condition are built from patches of the same image. Stimuli of the congruent and incongruent condition are built from patches of different images of the same class or of different classes, respectively. Permuted stimuli were created for each of the three conditions by shuffling the positions of bubbles.
Figure 11.
Each trial began with the presentation of a fixation point used for drift correction. Subsequently, the stimulus was presented for 3 seconds. The response screen was displayed until the subject responded to the classification task by pressing one of the indicated keys. The subject's choice was then shown as feedback.
Figure 12.
Computation of stimulus dependent salience.
For each bubble, stimulus dependent salience was computed by considering the luminance and texture contrast map of the embedding full field (A and C). Luminance and texture contrast at the location of the bubble (marked by red circles for one example bubble) are then mapped to fixation probabilities (red dots). These mappings (B and D) map luminance and texture contrast bins (see text) to fixation probabilities and were obtained in a baseline study using a large number of stimuli from different categories. The resulting fixation probabilities based on luminance and texture contrast were multiplied yielding the stimulus dependent salience.
Figure 13.
Simulation of fixation trajectories based on spatial biases.
Spatial bias salience was computed from simulated fixation trajectories based on the central bias of fixations, saccade statistics, and bubble positions. Given the current fixation location, the next fixation is generated by, first, multiplying the central bias map (A) with the bubble position map (B). Second, the resulting intermediate map (C) is multiplied with the probability distribution over saccade vectors (D) centered at the current fixation. The next fixation is then sampled from the resulting map (E). For example, assuming a fixation of the upper left bubble in panel C, the multiplication (indicated by the white coordinate frame) of the intermediate map (C) and saccade statistics (D) results in the depicted next fixation map (E). Repeating this sampling procedure resulted in the simulated fixation trajectory.