“…One of the earliest works on face recognition that describes a face with its within-individual and between-individual variations was introduced in [26] [27] [28]. Probabilistic Linear Discriminant Analysis (PLDA) [42] was employed to establish a generative linear model, and the optimal latent identity variable was iteratively derived by using the Expectation-Maximization (EM) [29] algorithm. This method was further applied to age-invariant face recognition in [21], where the within-individual variance was suitable for using the aging information, while the between-individual variation was suitable for using the identity information.…”