Rapid prototyping (RP) techniques have revolutionized traditional manufactu
ring methods. These techniques allow the user to fabricate a part directly
from a conceptual model before investing in production tooling. However, si
gnificant time is required to complete the geometric modelling stage. This
paper describes the development of an image-based rapid prototyping approac
h which can be used to generate geometric models in reverse design applicat
ions. The method, which uses standard two-dimensional photography, integrat
es the Torrance-Sparrow illumination model and the photometric stereo techn
ique. A neural network approach is employed to decide the illumination para
meters and the solution of the model. A range of examples has shown this me
thod to be feasible and efficient to support the RP applications.