What’s wrong with this picture? an experiment in quantifying accuracy in 2D portrait drawing

Abstract

Traditional portraiture aims to produce a life-like representation of an individual’s unique facial features, but there are a number of perceptual factors that may affect how an artist sees and depicts a sitter’s facial shapes. The anatomical accuracy of a portrait is predominantly ascertained subjectively, and not through quantitative methods. To see if it was possible to apply the methods of analysis more typically used to study variation in biological forms, the authors produced 30 portrait drawings based on 30 pose photographs, and analysed how the portrait drawings differed from the photographs using three different measures: visual assessment, anthropometric measures, and geometric morphometric analysis. Of the three methods, geometric morphometric analysis was found to be the far better technique to account for patterns of depiction common to one artist, and our study indicates this method may have broad research applicability to the face as it is variously represented across a range of visual media.

Cain’s house task revisited and revived: Extending theory and methodology for quantifying drawing accuracy.

“…(Hayes & Milne, 2011, p. 151) Hayes and Milne (2011) found that visual assessment did not coincide with anthropométrie morphometric evaluations of pitch and cant of the head. This suggests that subjective perceptions of the actual visual features of the portraits were inaccurate.…”

A geometric morphometric evaluation of the Belanglo ‘Angel’ facial approximation

“…All images were then visually assessed for head pose, face shape and facial expression. Head pose was estimated by the position of the ears, which can appear raised in a downwards head pitch and obscured in a head turn [26,28], face shape according to whether it appeared short/wide, tall/thin or an indeterminate combination, and facial expression by the shape of the lips and cheeks, which appear raised during smiling [48,49] [50]), indicating that the right turned faces tend towards a downward pose and vice versa. Visual assessment of faces by one individual is, however, known to be inaccurate with regards to estimating head pose, and requires a minimum of 10 assessors to statistically correlate with the accuracy of a geometric morphometric analysis [26,28].…”

“…While neither seems to display an upwards head orientation, both appear to have been taken from an angle above head height, which, on the basis of past forensic photographic comparisons using geometric morphometrics [e.g. 32,[33][34][35], results in a facial morphology that shares some of the shape changes resulting from a downwardly pitched head pose (the upper face is expanded and lengthened while the lower face is contracted and foreshortened [26]). One photograph is of comparatively low resolution and depicts Ms Pearce-Stevenson fairly frontally orientated with a relatively neutral/serious facial expression (Figure 1d), while in the higher resolution photograph her head is turned to the right shoulder (both eyes are visible but the right ear is obscured) and Ms Pearce-Stevenson is displaying a closed-mouth smile (Figure 1e).…”

“…However, and as mentioned previously, these independent variables are derived from inherently inaccurate visual assessments, and on the basis of past research findings [26,28] can be assumed to under-represent the statistical significance of the related shape variance.…”

“…Because of its capacity to process complex patterns of shape variance, previous research has found that geometric morphometrics can enable statistically meaningful comparisons between (i) highly disparate image types depicting the face of the same individual [25,26], (ii) widely divergent palaeoart facial reconstructions of the same skull [27], and (iii) between images of different individuals displaying a range of habitual head pose variations [28]. It is because of this ability to identify shape variance due to head pose that geometric morphometrics has been applied for this evaluation of the facial approximation of Ms Pearce-Stevenson, though photographic distortion remains a confounding variable.…”