Can AI-assisted objective facial attractiveness scoring systems replace manual aesthetic evaluations? A comparative analysis of human and machine ratings.

BACKGROUND

In clinical practice, attaining a genuinely objective evaluation of facial aesthetics has posed considerable challenges owing to the inherent subjectivity of human observers. Artificial intelligence (AI) technology has demonstrated significant potential in facial aesthetic evaluation owing to its inherent objectivity. However, whether it can objectively reflect public aesthetic preferences and its consistency with human aesthetic evaluations remains controversial. This study aimed to explore the consistency between AI facial aesthetic scores and human aesthetic preferences, evaluating its potential and reliability in clinical applications.

METHODS

The study used the SCUT-FBP5500 facial aesthetic dataset and employed the Face++ facial analysis algorithm to score facial aesthetics. Sixty volunteers provided manual aesthetic scores for the dataset. Correlation coefficients were used to assess the relationship between scores. Intraclass correlation coefficients (ICC) were used to evaluate the consistency between AI and manual facial aesthetic scores.

RESULTS

AI scores showed a good overall correlation with manual scores (r = 0.778, p<0.001). The average correlation coefficient between AI and each individual rater (r = 0.60 ± 0.06) showed no significant difference compared to the average level among raters. Consistency analysis indicated moderate consistency between AI and manual scores (ICC = 0.66 ± 0.10). The AI system demonstrated high accuracy in classifying high and low facial aesthetic scores (accuracy = 0.788, precision = 0.792, recall = 0.786, F1 score = 0.789).

CONCLUSION

The AI facial aesthetic scoring technology demonstrated good application potential, with high correlation and consistency with human raters. It is expected to serve as an objective aesthetic evaluation tool for clinical aesthetic assessments.