In order to avoid the generation of blurred images principle component analysis models (21) are generated by identifying points in the images outlining facial features in the images and creating a triangulation of the face of an individual in which boundaries of the facial features of an individual lie substantially along the edges of triangles in the triangulation. A principle component analysis model (21) is then created using the variation in coordinates of the points forming the triangulation and the differences in image data of the images divided by the triangulations. As the edges of facial features tend to lie along edges of the triangles, morphing of these shapes does not cause these features to be stretched across two triangles and hence variation in facial features such as the shape of eyebrows or eyes is primarily encoded in the variation of location of the vertices of the triangulation rather than differences in the generated morphed images used to determine coloring for generated models.