Optimized facial landmark modeling with medical aesthetic constraints by a multi-objective genetic algorithm

Area of Science:

• Computer Vision
• Medical Aesthetics
• Facial Analysis

Background:

• Facial beauty is a subjective construct influenced by social and cultural factors.
• Quantifying facial attractiveness requires integrating computer vision and medical aesthetics (MAs).
• Previous methods lacked professional judgment and theoretical basis for beauty assessment.

Purpose of the Study:

• To develop a professional and theoretically-grounded method for facial beauty assessment.
• To enhance the accuracy of facial beauty prediction using MA features.
• To introduce a novel feature selection algorithm and facial landmark model for precise scoring.

Main Methods:

• Derived MA features encompassing global, local, and curvature aspects based on aesthetic principles.
• Developed an aesthetic-driven feature selection algorithm within a multi-objective evolutionary framework.
• Introduced an MA facial landmark model for precise annotation of key facial points.

Main Results:

• Achieved superior performance on SCUT-FBP, SCUT-FBP5500, and Chicago Face Dataset.
• Reported high accuracy with Pearson's correlation coefficient = 0.8216, MAE = 0.2638, RMSE = 0.3743.
• Validated the clinical relevance and practical applicability of the proposed method.

Conclusions:

• The study provides a practical tool for beauty evaluation aligned with professional judgments.
• The developed method enables transparent and explainable outcomes in clinical and cosmetic applications.
• This approach enhances predictive accuracy in facial beauty assessment.