's, that
could signal a possible silhouette in the Mesh Method and
Tessellated-Mesh Method. Once it is determined that there is a
silhouette on the control mesh and hence a possible silhouette on the
surface, there would need to be a check to determine whether or not
there is a silhouette on the surface.
The Srf-Node method is the most concise and does not require the maintenance of large data structures which may bog down the desired interactive rates. However, the Srf-Node method has the restriction that all of the surfaces must be sufficiently refined. Although this increases the number of dot products that need to be calculated, most of operations can be done as a preprocess.
In order to achieve real-time geometric silhouette calculations, one could take advantage of previously calculated silhouettes and the movement of the view to approximate the next set of silhouettes. As in Markosian's algorithm, there would have to be some way of determining where silhouettes may suddenly appear, i.e., a region that suddenly has silhouettes that were not in the previous frame. Normal or visibility cones [33,20] could also be used to rule out where silhouettes could not be; then a test of the whole surface may not be necessary. If the model can be sufficiently tessellated, Markosian's algorithm may be preferred. However, for a highly trimmed NURBS model, tessellation may not be the best choice.