DFDnet: Discriminant Face Descriptor Network for Facial Age Estimation

Age estimation is an important part of biometric recognition. In this paper, we propose to use Discriminant Face Descriptor (DFD) which learns the most discriminative and related features to age variation in a data-driven way. We use it, for the first time, for facial age Age estimation is an important part of biometric recognition. In this paper, we propose a stacked structure called Discriminant Face Descriptor network (DFDnet) to learn the most discriminative and related features to age variation. We extract the multi-stage Discriminant Face Descriptor (DFD) features which are age-sensitive. We first introduce DFD for facial age estimation instead of original face recognition. Then the bag-of-features method is used so that each image can be represented by the histogram of visual words. Finally, age estimation is achieved via simple linear regression algorithm. Experiments on the publicly available MORPH and FG-NET databases validate the effectiveness of our proposed DFDnet method. To further illuminate the usefulness of our approach in unconstrained environments, we also conduct the experiment on Cross-Age Celebrity Dataset (CACD) which is collected from Internet movie database (IMDB).