3D Face Cloning from orthogonal photographs

    It utilizes a modification of a generic model after detection of feature points on picture data and generate texture mapping including fully automatic texture image generation. The methods for cloning are able to make 3D animation of a face from orthogonal picture data in a few minutes (practically about 1~5 minutes).
    2D photos offer clues to the 3D shape of an object. It is not feasible, however, to consider 3D-points densely distributed on the head. In most cases, we know the location of only a few visible features such as eyes, lips and silhouettes, the key characteristics for recognizing people, which are detected in a semiautomatic way using the freeform deformation and then the structured snake method with some anchor functionality for a subset of feature points, say key features. Figure 1(a) depicts an orthogonal pair of normalized images, showing the detected features. The two 2D sets of position coordinates, from front and side views, i.e., the (x, y) and the (z, y) planes, are combined to give a single set of 3D points. The problem is how to deform a generic model, which has more than a thousand points to make an individualized smooth surface. One solution is to use 3D feature points as a set of control points for a deformation. Then the deformation of a surface can be seen as an interpolation of the displacements of the control points. In the Dirichlet-based FFD approach, any point of the surface is expressed relative to a subset of the control points set with the Sibson coordinate system. Therefore DFFD is used here to get new geometrical coordinates for a modification of the generic head on which are situated the newly detected feature points. The correct shapes of the eyes and teeth are assured through translation and scaling appropriate to the new head. As shown in Figure 1(b), the result is quite respectable when we consider the input data (pictures from only two views). To improve the realism, we make use of automatic texture mapping with texture generation together.

      Figure 1: (a) Modification of a generic head according to feature points detected on pictures. Points on a 3D head are control points for DFFD. (b) Snapshots of a reconstructed head in several views.
    Texture mapping serves not only to disguise the roughness of shape matching determined by a few feature points, but also to imbue the face with more realistic complexion and tint. If the texture mapping is not correct, the accurate shape is useless in practice. We therefore use information from the set of feature points detected to generate texture in a fully automatically, based on the two views. The main criterion is to obtain the highest resolution possible for most detailed portions. We first connect two pictures along predefined feature lines using geometrical deformations and, to avoid visible boundary effects, a multiresolution technique. We then obtain appropriate texture coordinates for every point on the head using the same image transformation. Figure 1(b) shows several views of the head reconstructed.

WonSook LEE, Ph.D.

School of Information Technology and Engineering (SITE)
University of Ottawa
800 King Edward Avenue
P.O. Box 450, Stn A
Ottawa, Ontario, Canada K1N 6N5

Office: SITE, Room 4065
Tel: (613) 562-5800 ext. 6227
Fax: (613) 562-5664
Email: wslee@site.uottawa.ca
Web: http://www.site.uottawa.ca/~wslee