Recurrent Age Recognition Based on Manifold Learning

With the rapid development of big data and artificial intelligence, facial age recognition has become an active and hot research field. Existing age recognition often focuses on facial features from single facial images, without taking into account personal aging patterns. We propose an algorithm for personal aging patterns based on manifold learning (MLPAP). MLPAP combines manifold learning and two deep learning architecture Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU). CNN is trained to extract facial features, manifold learning embeds high-dimensional facial features extracted by CNN into low-dimensional discriminative subspaces, and GRU is employed to learn personal aging patterns from the sequential individual features. Furthermore, the improved label distribution learning (LDL) scheme integrated into MLPAP to exploit ambiguity from the biological and adjacent ages. The performance of the MLPAP is evaluated, and the results show that MLPAP has the best performance compared with the state-of-the-art methods.