Reconstructing 3D Face of Infants in Social Interactions Using Morphable Models of Non-Infants

Area of Science:

• Computer Vision
• Developmental Psychology
• Biomedical Imaging

Background:

• 3D morphable models (3DMMs) are crucial for facial reconstruction but are primarily developed for adults.
• Infant 3DMM fitting is an under-researched area, with limited understanding of adult model applicability to infants.
• Accurate facial analysis in infants is vital for understanding early development and behavior.

Purpose of the Study:

• To evaluate the efficacy of state-of-the-art 3DMM fitting methods for infant facial reconstruction from 2D images.
• To determine if existing adult-centric 3DMMs can be reliably applied to infant data.
• To explore the potential of 3DMMs for analyzing infant facial behavior.

Main Methods:

• Comparison of five advanced 3DMM fitting techniques on naturalistic infant-caregiver interaction videos.
• Analysis of reconstruction accuracy using single-frame versus multi-frame data.
• Qualitative assessment of facial region reconstruction, focusing on areas with high texture variation.

Main Results:

• Consistent, subject-specific 3D facial identity reconstruction is achievable from multiple frames, but not from single frames.
• Facial regions with high texture, such as eyes, brows, and mouth, are reconstructed with greater accuracy.
• Adult-developed 3DMMs exhibit limitations in reconstructing the entire infant facial morphology.

Conclusions:

• While adult 3DMMs have limitations for full infant facial morphology, they are feasible for infant facial behavior analysis.
• Multi-frame 3DMM fitting shows potential for consistent infant facial identity and behavior tracking.
• Further research into 3DMMs for infant facial behavior measurement is warranted, leveraging their ability to mitigate identity and pose variations.