V-FACE: A Large-Scale Vietnamese Face Image Database in Unconstrained Environments

Abstract

Researches have been conducted to exploit and collect details in every part of the human body to understand ourselves and improve many aspects of science and technology. Especially, the face is one of the most informative parts that is used commonly in computer vision such as face identification, face recognition, and landmarks detection, etc. A number of face image databases have been published and used worldwide in different technologies. In this paper, we introduce a new large-scale face image database for Vietnamese and describe a capturing system specifically designed to obtain the data. The V-FACE database contains more than 3 million high-quality images of more than 300 subjects and has a balanced ratio of genders. The database includes a variety of attributes, including different angles, lighting conditions, facial expressions, and occlusions with the combination of four types of accessories. Therefore, the V-FACE database can be exploited to be used in various tasks such as unconstrained face recognition, 3D face model reconstruction, face image frontalization, etc. In a variety of unconstrained environments, V-FACE provides a sufficient amount of data to improve the accuracy of face recognition technologies in practical scenarios. Additionally, with the information on different poses, we can reconstruct the human face in the most accurate details. There are several other Vietnamese face image datasets such as COMASK20 and VN-Celeb. V-FACE is a large-scale Vietnamese face database with a balance ratio in genders and age groups and different unconstrained environments such as illumination, facial expressions, poses and occlusions. We hope this database will improve many types of research in the world and especially in Vietnam.