Abstract:
Realistic images of faces of individual people have been generated by computer by texture mapping a photograph of the individual onto a wireframe model of their head. In previous work each individual had to be measured in order to build this model.
We use a generic head model to describe the shape of a human head. This model gives us ``world knowledge'' of the shape of a head, to create the new images we distorted it to fit the face in the photograph, creating a new, specific, head model. This specific model can be distorted further to change the expression before it is projected onto a viewing plane. We then map the texture from the photograph of the person, one triangle of the model at a time, onto the viewing plane to create a novel view of the individual. Texture from more than one image may be used, by selecting the most appropriate image for each triangle of the model.
The generic --> specific transformation is implemented by making each edge of the model into a spring, displacing some of the vertices to fit the photograph and then allowing the spring-frame model to relax to a position of least energy.
Key Phrases: computer graphics, image warping, texture mapping, multiple texture sources, synthetic images, wireframe model, generic model, specific model, spring-frame model, human heads, human faces.