Machine Learning Generated Synthetic Faces for Use in Facial Aesthetic Research.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Medical Imaging

Background:

• Facial aesthetic research requires extensive, accessible datasets.
• Current datasets often have usage restrictions, hindering research.
• A need exists for a centralized, unrestricted repository of facial images.

Purpose of the Study:

• To generate a repository of synthetic facial images using machine learning.
• To analyze these synthetic faces based on established aesthetic principles.
• To assess the utility of synthetic images as a research tool.

Main Methods:

• Utilized an open-source generative adversarial network (GAN) to create synthetic facial images.
• Employed computer vision technology for image refinement and analysis.
• Extracted key facial attributes such as age, gender, emotion, and landmarks.

Main Results:

• Generated 1,000 synthetic facial images, with 60 analyzed in detail.
• Synthetic images adhered to aesthetic principles like horizontal thirds and vertical fifths.
• Demonstrated high correspondence between synthetic and real human facial photographs.

Conclusions:

• Successfully generated realistic synthetic facial images.
• The synthetic dataset shows potential as a valuable research tool.
• These AI-generated faces align with aesthetic principles and resemble real photographs.