![]() |
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ИНХС РАН |
||
The prediction of a person’s chronological age (CA) from visible skin features using artificial intelligence (AI) is now commonplace in computer vision. Most often, this is accomplished by convolutional neural network (CNN) models based on facial images as the biometric data. However, hands also hold telltale signs of a person’s age, maybe even more telling than the face. To determine the utility of hand images in predicting CA, we developed two deep CNNs, one based on standardized dorsal hand images (H) and another on frontal face images (F). Subjects (n=1454) were Indian women, ages 20-80, across 3 geographic cohorts (Mumbai, New Delhi, and Bangalore). Given the wide variability in skin types, each subject’s individual typology angle (ITA), a measure of overall skin color tone, was determined on both F and H. Images were randomized and 70% of F and 70% of H were used to train the two CNNs. The remaining 30% of F and H were retained as the validation dataset. CNN validation showed mean average error (MAE) for predicting CA using F vs. H of 4.1 and 4.7 years, respectively. Furthermore, CA predicted from F was highly correlated with that predicted by H (r=0.96, p<0.0001). The CNNs for both F and H were validated for dark and light skin color tones. Finally, by blurring or accentuating specific visible features on regions of the hand and face such as wrinkles and spots, we identified those features that contributed more or less to the CNN models. For the face, areas of the inner eye corner and areas around the mouth were most important for age prediction, however with some age dependence. For the hands, knuckle texture was a key driver for age prediction. In conclusion, for AI estimates of CA, CNNs based on hand images are a viable alternative and comparable to CNNs based on images of the face.