Face Aging Using Generative Adversarial network

Abstract

If we provide someone with a large amount of faces in different ages which is not paired and asked him if he can generate a photo of his face at older or younger age his answer would be probably no. Providing someone with an older or younger face of him is a complex task. Nowadays most face aging models are required paired sample as the dataset contains someone face and an older face of the same person to learn the traversing between the two age. Collecting a paired samples dataset is a complex task to achieve and have an availability limitation. In this paper, the model did not require a paired sample it only require a dataset for every group of age from 0 – 100 (group for each ten years). The model learns the face details between each group of age. In addition, that the model not only generate an age progression and regression faces, but it also saves the personality of the face during the aging process. The face personality is saved during aging according to the use of encoder which down sample the image to save the high-level face feature. The model structure consists of encoder and generator to generate progression and regression faces. Also, it consists of two adversarial networks respectively to force generating real faces.