Deep Age Estimation Model Optimization

Face recognition is the predominant biometric method to identify and verify human facial features due to its stability in determining a person’s identity. However, face recognition models can be improved relative to several challenges, e.g., the reliability of recognizing faces at different ages. Facial aging, which affects the shape and texture of the face, occurs in all individuals. Thus, this study attempted to optimize the performance of neural network-based age estimation methods by modifying the network architecture and hyperparameter values. In the future, we expect that the face recognition architecture will be employed to construct several models based on different age ranges. Consequently, the performance of the face recognition model within the specific age ranges is optimized. The modifications applied in this study resulted in a mean absolute error (MAE) value of 5.40 during validation for age estimation, indicating that the model estimated age more accurately than before, where MAE values of 7.60 and 5.53 were obtained.