Synthesis and transformation of three-dimensional facial images

In this article, we enhance and extend 2-D methods by using texture-mapped laser-video depth data. Methods for the synthesis of new three-dimensional (3-D) facial images from a database of texture-mapped laser-scanned surfaces are described. Four methods are used to create new facial images. The first constructs average facial images from a sample of faces. The second exaggerates the difference between a subject and an average to create a caricature. The third method captures the difference between two averages, which can then be applied to individuals for statistical transformation. Finally, principal component analysis is used to find the major axes of facial variation in a sample, allowing the construction of completely novel faces from a small vector of values.