An Automatic Facial Age Proression Estimation System

Linear age progression models which are largely used in prototype and conventional approaches usually produce synthesized images that are lack of quality because of the aging variations. Therefore, in this paper, a facial age progression model that captures non-linear age variances is designed by using a deep learning-based method called Generative Adversarial Network. The proposed face aging model aims to achieve convincing and visually plausible aging effects by controlling the age attribute. The model first maps the face via a convolutional encoder to a latent vector, then the vector is projected by a deconvolutional generator to the face manifold based on age, and finally the encoder and the generator are imposed on two adversarial networks respectively. The proposed model is trained on UTKFace dataset using Pytorch machine learning library. The experimental results demonstrate the capability of the proposed Generative Advanced Network (GAN) model of generating photorealistic aging faces and preserving the original identity property.