Fig. 5: Reliability of the first principal component (PC1) across image categories.
From: Visual exploration dynamics are low-dimensional and driven by intrinsic factors
The full set of features used for the principal component analysis (PCA) in the main analysis was extracted separately for each image category (i.e., indoor, outdoor natural, outdoor manmade, scenes with humans, scenes without humans). Next, we computed category-specific individual PC1 scores in the component space of the main PCA by applying PC1 loadings (calculated on all images) on features computed from each category. This procedure compares PC1 scores obtained on all images vs. specific image category. The upper five scatterplots represent the correlation between PC1 scores extracted from each specific image category (x-axis) and from all images (y-axis). The similarity is very high (all Pearson’s r = 0.97). The matrix shows the PC1 scores for each subject across different image categories. Note high variability across subjects and similarity across image categories.
