Avatar Recommendation Method Based On Facial Attribute

Abstract

An avatar is a representation of a real person in a virtual world and a user considers it as his/her secondary personal identity. With the spread of Virtual Reality(VR) technologies, there has been a user’s growing interest in how to make an avatar to describe his/her identity well. Recent commercial games also support powerful avatar customization services to satisfy customer’s various needs, but it is not easy to create a satisfactory avatar because there are typically dozens of customization options. Thus, we argue that a proper recommendation system is necessary for a good avatar customization service to support customer’s choice. Customizing an avatar is affected by a variety of aspects such as personal tastes or an avatar’s characteristics. For example, a user can build an avatar whose sex is opposite or whose species is different from him/her-self. This is the reason why simple template matching or some similarity-based methods are not enough for avatar recommendation. In this paper, we propose a facial attribute-based avatar recommendation method where avatars are recommended from a face picture and preferences given by a user. A face plays an important role in a personal identity, and therefore abstracting a face by its attributes helps deal with a diversity of customization requirements such as different sex or species.