Institute of Computing for Physics and Technology
Protvino, Moscow region, Russia
(Updated version of the paper presented on Graphicon 2010)
The done tests demonstrate that further increasing of the mesh repairing quality and robustness is possible. But it requires using more and more costly missing surface estimation methods. So, parallelization of computation is only way to increase the mesh repairing quality keeping acceptable performance indices. The introduced AF concept provides good potential for development new missing surface estimation methods and in the field of parallelization as well. As a particular, it seems perspective to supplement the done implementations of AF by an implementation that uses some database of samples and a neural network. The “physical” style of the AF concept corresponds to the abilities of modern GPU. Using them promises a great benefit.
The method has the proved ability to repair heavily damaged triangle mesh models. In general, the mesh repairing can be performed without any manual management. At the same time, fine adjustment of some parameters (for example, the distance function factors) allows to obtain a better result. Considering the mesh repairing quality as a function of these parameters, such adjustment can be made using global optimization methods on PC-clusters.