Image-Based Age
Description
Image-based age estimates an individual’s age (in years) based on information obtained from video (or a single image) of their face.
There are several types of image-based age:
Topographical Age – an estimate of age based on features of the entire image, including face shape, wrinkles, and skin condition. This measure is currently displayed as “Facial Skin Age” in the BitDoctor app. It can be influenced by factors such as fatigue and the use of skincare and cosmetic products. Video images taken in poor lighting conditions (e.g., backlighting, overhead lighting) may distort this estimate.
Physiological Age – an estimate of age based on blood flow information acquired with BitDoctor Technology. It can be used as an indicator of the effects of aging on your facial vasculature.
Total Facial Age is an estimate of age that combines information from 2 types of imagebased age (topographical and physiological).
Participants
Topographical Age Model – dataset of images and demographic information from 1.6 million individuals. This is a multiracial dataset consisting mostly of North American participants.
Physiological Age Model – video of the face and demographic information collected from 14,970 participants.
All participants were adults 18+ years of age.
Additional Data Collectioon Procedure
A well-lit single image photograph of the face was taken for both models.
Modeling Approach
Deep learning models were trained to predict the subject’s actual age from various image types, depending on the model. The topographical age model was trained using single images of the face, with each image coming from a unique individual. The physiological age model was trained using blood flow features extracted from 30s of facial video and averaged over the duration of the video.
Model Performance
Image-based age models were validated on an independent dataset that was not used in training (data was collected in China akin to that used for physiological and Visage age models). The distribution of sex and age in this group is summarized in Table 27.
Sex distribution
49.4% Male; 51.6 % Female
Age
53.87 ± 14.40
Table 27: Age distribution of pristine validation set
Model performance for the individual components of the model is summarized in Table 28. Combining information from all three models (Total Facial Age) results in the best performance.
Topographical Age Model
0.77
-2.87 ± 6.04
Physiological Age Model
0.77
-3.96 ± 9.25
Total Facial Age (Topographical + Physiological)
0.94
-4.11 ± 4.77
Table 28: Facial skin age model performance
Last updated
