Signatures of personality on dense 3D facial images

Sile Hu,Jieyi Xiong,Kun Tang,Lu Qiao,Pengcheng Fu,Li Jin,Jingze Tan

doi:10.1038/s41598-017-00071-5

Abstract

It has long been speculated that cues on the human face exist that allow observers to make reliable judgments of others’ personality traits. However, direct evidence of association between facial shapes and personality is missing from the current literature. This study assessed the personality attributes of 834 Han Chinese volunteers (405 males and 429 females), utilising the five-factor personality model (‘Big Five’), and collected their neutral 3D facial images. Dense anatomical correspondence was established across the 3D facial images in order to allow high-dimensional quantitative analyses of the facial phenotypes. In this paper, we developed a Partial Least Squares (PLS) -based method. We used composite partial least squares component (CPSLC) to test association between the self-tested personality scores and the dense 3D facial image data, then used principal component analysis (PCA) for further validation. Among the five personality factors, agreeableness and conscientiousness in males and extraversion in females were significantly associated with specific facial patterns. The personality-related facial patterns were extracted and their effects were extrapolated on simulated 3D facial models.

Highlights

As one of the most complex anthropological traits, the human facial shape is strongly regulated by many factors such as genetics, ethnicity, age, gender, and health
Thereafter, we proposed a partial least squares (PLS) based statistical method called CPLSC to inspect the association between the human face and each of the Big Five’ (BF) factors based on the 3D image data and the BF scores, and extract the personality-related facial features from the high-density 3D image data
We carried out PLS regressions for each of the five BF factors separately, and a leave-one-out (LOO) procedure was applied to cross-validate the PLS models

Summary

Introduction

As one of the most complex anthropological traits, the human facial shape is strongly regulated by many factors such as genetics, ethnicity, age, gender, and health. Passini and Norman first conducted a seminal study in 1966, in which a small group of volunteers, unknown to each other, were asked to rate themselves and their peers on the BF scales without verbal communication[30] They found that for extraversion, agreeableness, and openness, the self-reported scores and those scored by observers matched significantly. In order to directly obtain the specific facial patterns associated with personality traits, Little and Perrett proposed an ingenious method[33] They ranked the head portraits of the volunteers along the five BF dimensions based on their self-ratings, and synthesised the composite portraits for the extreme scorers. Thereafter, we proposed a partial least squares (PLS) based statistical method called CPLSC to inspect the association between the human face and each of the BF factors based on the 3D image data and the BF scores, and extract the personality-related facial features from the high-density 3D image data. All the extracted personality-related features were visualised and animated by our R package “3DFace”

Methods

Results

Conclusion