A separated linear representation of shape and texture for modeling appearance variations of human faces : toward image synthesis transforming impressions conveyed by human faces

Abstract

This paper describes a linear representation of the separated shape and texture spaces obtained by principal component analysis of the facial images that allows a smooth modeling of variations in the appearance of human faces. A computational model of image transformation acting on visual impression of the faces has been proposed by applying Fisher's liner discriminant analysis on the separated linear shape and texture spaces respectively, and rating evaluations are being made on its applicability to synthesis of impression-manipulated facial images.