An overview of facial attribute learning

Abstract

Facial attributes are useful for developing applications such as face recognition, search, and surveillance. They are therefore important for various facial analysis. Many facial attribute learning algorithms have been developed to automatically detect those key attributes over the years. In this paper, we have surveyed some typical facial attribute learning methods. Five major categories of the state-of-the-art methods are identified: (1) Traditional learning, (2) Deep Single Task Learning, (3) Deep Multitask Learning, (4) Imbalanced Data Solver, and (5) Facial Attribute Ontology. They included from traditional learning algorithm to deep learning, along with methods that assist in solving semantic gaps based on ontology and solving data imbalances. For each algorithm of category, basic theories as well as their strengths, weaknesses, and differences are discussed. We also compared their performance on the standard datasets. Finally, based on characteristics and contribution of methods, we present conclusion and future works to solve facial attributes learning. The survey can help researchers gain a quick overview to build future human face applications as well as further studies.