Improving Face Age Prediction by Using Multiple-Angle Photos.

Area of Science:

• Computer Vision
• Biotechnology
• Gerontology

Background:

• Face photo-based age estimation is a cost-effective tool for biological age studies.
• Current AI age models typically use single front-view photos, limiting potential accuracy.

Purpose of the Study:

• To investigate if using multiple facial photos from different angles improves AI-based age prediction.
• To compare the performance of age prediction models trained on front-view, side-view, and combined facial images.

Main Methods:

• Developed AI age prediction models using a dataset of mugshots.
• Trained models on single front-view images, single side-view images, and combined front and side images.
• Evaluated model performance using mean absolute error (MAE).

Main Results:

• Accurate age prediction was achieved with both front (MAE=3.1 years) and side (MAE=3.7 years) views.
• Combining front and side views yielded the best performance (MAE=2.88 years).
• Side-view models demonstrated better robustness to facial rotation compared to front-view models.

Conclusions:

• Using two facial photos from different angles slightly enhances age prediction accuracy.
• Multi-angle facial photos provide a more robust approximation of biological age than single photos.
• This approach can benefit personalized medicine, aging intervention, and rejuvenation studies.