PROSOPO-A Face Image Synthesis System

Abstract

In this paper we address the problem of synthesizing face images displaying prede.ned facial attributes, such as the shape and color of the overall face, the shape and color of individual facial features and the age of the subject shown in the synthesized image. The face image synthesis method is based on a statistical face model that enables the reversible representation of face images using a small number of parameters. By manipulating the parameters of the face model it is possible to generate di.erent instances of faces. We describe how the mathematical relationship between the model-based representation of faces and facial attributes of interest can be de.ned and used for transforming a set of facial attributes to model parameters, so that a face image consistent with the description provided can be synthesized. Our methodology can be applied to the problem of generating facial images of suspects.