The paper addresses the task of automated determination of the extroversion/introversion (E/I) dichotomy based on the analysis of static facial photographs; proposes and experimentally investigates an algorithm based on extracting facial geometric features using the MediaPipe FaceMesh framework and subsequent classification with the K-Nearest Neighbors (KNN) algorithm. The relative distances between key anatomical facial points are used as features, which minimize the influence of image scale and shooting conditions. The empirical basis of the study is an author’s dataset formed based on respondents’ photographs with verified test results using the MBTI methodology. The authors conduct a series of experiments with varying sets of features, classifier parameters, and image selection criteria. The work shows that the quality and standardization of photographs (frontal view, absence of facial expressions, makeup, and head turns) have a critical impact on recognition accuracy. The maximum classification accuracy achieved is approximately 72% on the selected sample. The results obtained confirm the statistically significant relationship between the morphological characteristics of the face and the E/I dichotomy.
Polozov et al. (Tue,) studied this question.