The Pull-Push Algorithm Revisited

Martin Kraus
Computer Graphics and Visualization Group, Technische Universität München, Germany

Background

The pull-push algorithm is a well-known and very efficient pyramid algorithm for the interpolation of scattered data with many applications in computer graphics. However, the original algorithm is not very well suited for an implementation on GPUs (graphics processing units). In this work, several improvements of the algorithm are presented to overcome this limitation, and important details of the algorithm are clarified, in particular the importance of the correct normalization of the employed filters. Moreover, we present an extension for a very efficient estimate of the local density of sample points.