Automatic avatar creator

Abstract

The proposed system aims to recreate a user digitally as an avatar, taking a live webcam feed as input. This digital caricature will mainly focus on the head and translate features such as skin colour, hair style, hair and eye colour together with mouth style and any eye wear detected. It aims to offer a less time consuming approach to avatar construction and provide a solid starting point for further customizability. All that is required to run the program is a windows pc and a standard webcam. After the installation, the program launches automatically and a webcam feed may be seen on the left whilst an avatar editor is seen on the right. If the user is positioned at an adequate distance from the webcam and the room lighting is not of a hindrance, the face will be detected instantly through the use of appearance-based detection methods employed in the Open CV library. The same method is then employed to detect eyes, nose and mouth and in this way feature extraction may take place. Extracted readings may be classified as either 'hard variables' or 'soft variables'. The former are only detected once and do not change throughout the whole session. On the other hand 'soft variables' such as glasses and 'mouth style' detection are constantly being refreshed and should reflect real-time changes. In other words once the hair colour is detected as "black" it will not change throughout the whole session. However, putting on a pair of glasses after the avatar has been constructed will result in a change. The outcome of this process can be seen in the avatar window where the detected features will be reflected in the digital model. The editor is also open to manual changes, giving the user the ability to correct wrongly detected features or else change other characteristics which for some reason or another were not automatically detected. From the results obtained, the proposed solution seems to be fairly successful in obtaining the correct facial information and delivers on its promise to link features extraction to avatar construction.