I already explained in my comments to the question that there is no enough information to build a 3D model, ill-posed problem like in tomography, etc. Even a human eye do not really reconstruct such mode. You would rather have a set of photographs at different angles. In 1999, I learned from some MetaCreation's people about their camera which used only two shorts, but one of them was made using specially structured lighting. Similar structured-lighting approach is used in some pseudo-3D detectors used to build a 3D profile of the Printed Circuit Board in electronic industry. Sorry I cannot give much more detail, but those approaches are very interesting and require very special hardware. This is an interesting article about building 3D models from photos:
http://www.pcworld.com/article/11311/build_3d_models_from_photos.html[
^].
My little 2D ideas are rather trivial but seem realistic to me. First, I would recommend you to load and use Inkspace:
http://en.wikipedia.org/wiki/Inkscape[
^],
http://inkscape.org/[
^].
Please run it and play with it. It could be recommended as a primary tool for developing all kind of graphics for software, but not just for software. There are many different effects supplied with this wonderful vector editing tool, but I would like to bring your attention to two of them: [Main Menu] => Path => Trace Bitmap and [Main Menu] => Path => Simplify. These two effects can help to vectorize the photograph and morph it give it some cartoonish look. Note, that simplify actually enriches the image, "simplification" simply means the process reduces the number if curve's nodes, but it gives it more sophisticated look. You also can search for effects of "posterization" (elsewhere, please see
http://en.wikipedia.org/wiki/Posterization[
^]). As Inkscape is Open-Source software, you can locate the source code of the effects and learn how to implement them.
My other more advanced but more fuzzy idea is the generalization of posterization. In principle, it's possible to decompose an image into a set of areas characterized with some "dominating gradient". The gradient withing each area can be reconstructed by interpolation and added to the model. So, the final model of the image should present a set of areas bounded by Bezier curves with gradient fill in each of them. Some problem could be smooth stitching them on their boundaries. I feel that the most adequate approach would be… ignoring this problem. Depending on how decomposition into areas is done, this effect can create visible color boundaries between the areas, like in classical posterization. It could support cartoonish look of the image and look natural enough.
After all, the whole field of painting including portrait painting is about the play with human perception. In most graphical styles, a human head is depicting with the use of some strokes on the face, curved lines with some thickness. There are no such line in real life! Nevertheless, the artistic styles of drawing often show the lines; and human perception reconstructs 3D image of a human head from this non-existent-in-nature images.
—SA