Learning to draw parametric faces

Humans' ability to draw faces accurately from memory is extremely rare. One source of difficulty is the drawing process itself, which requires converting a complex, three-dimensional mental representation to a two-dimensional drawing. To simplify the drawing process and more directly assess people's recall of faces, we used the Parameterized Face Drawing (PFD) model (Day & Davidenko, Visual Cognition, 26(2), 89-99, 2018; Day & Davidenko, Journal of Vision, 19(11):7, 1-12, 2019) to generate simplified face stimuli that non-artists could draw. In Experiment 1, participants (N = 37) completed a sequential drawing-by-copying task in which they were given 60 s to draw each of 18 target faces by copying them using a stylus on a touchscreen. Following each drawing, participants were prompted to label a set of 20 key points on their drawing based on a reference face, which were used to compute the drawing's accuracy. In Experiment 2, participants (N = 22) observed each target face for 15 s and were then given 60 s to draw it from memory. In Experiment 1, the accuracy of drawings improved slightly over the course of the 18 trials, although most of the improvement occurred during the first few trials. In Experiment 2 (drawing-from-memory), there was no evidence of improvement, although the null results are tentative given the small sample size. Despite weak evidence of learning, participants' drawings captured the likeness of the target faces significantly better than chance. We discuss implications of these findings for the use of drawing as a method of face recall.