Fig 1.
Left: Images similar to these were use as practice stimuli to familiarize participants with the tracing task and program interface. Images under Creative Commons Licence by Supachoteamorn, Aryadna Pons, Ksanyi, and Christopher Michel. Right Experiment setup with a pen display graphics tablet. The computer screen displayed instructions and the original 800x600 pixels photograph as reference during the trial. The graphics tablet displayed a 1440 x 1080 pixels copy of the image with a semi-transparent overlay for tracing. The Next, Finish, Show, and Undo functions corresponding to each button on the side of the graphics table were displayed during the entire experiment. Images by the authors. The set of 18 photographs are available at: https://osf.io/x9uj5/.
Fig 2.
(A) Ranking of line drawings was performed by re-arranging images printed on cardstock on a large table. Participants were instructed to first rank five randomly chosen drawings, and to then insert subsequent drawings at the appropriate position. (B) Constructing a super-reference. The highest-ranked drawing served as the initial reference. The other drawings were matched to the reference in ranking orders. Unmatched contours were added to the reference, eventually yielding a super-reference that contains the super set of contours from all drawings. Original photograph under Creative Commons License by Flickr user Cyndy Sims Parr.
Fig 3.
Super-references for all 18 scene images.
Table 1.
Inter-rater agreement of line drawing rankings.
Fig 4.
Average rank of drawings versus the ranks predicted by the multiple linear regression model.
The regression line is shown in red with the 95% confidence interval in gray.
Table 2.
Regression table for the regression in Eq 1.
Fig 5.
(A) Construction of half drawings according to consistency scores. The two drawings did not share any contours, and each contained approximately 50% of total pixels. (B) Experimental procedure for the six-alternative forced-choice scene categorization experiment.
Fig 6.
(A) Most consistent half-drawings. (B) Least consistent half-drawings.
Fig 7.
(A) Categorization accuracy for least and most consistent half drawings. (B) Number of pixels in particular types of contours in least and most consistent half drawings. ** p < 0.01, *** p < 0.001.