The aim of the thesis is the implementation and comparison of algorithms for reconstructing 3D models from the segmented volumetric data. Multi-level partition of unity approach for generating weighted local implicit functions is implemented to existing program for visualising 3D models, NeckVeins. These functions are used to approximate local behaviour of input points. 3D model is then created using exhaustive Bloomenthal polygonization. The algorithm is optimized with the use of k-d trees for finding nearest neighbours and with support for parallel operation on multiprocessor systems. The result is a Java program with a variety of different parameters which control the accuracy, speed and smoothness of the final 3D model. Instead of a direct volume polygonization with static resolution, MPUI creates an implicit function that can be rasterized with any algorithm for implicit surface polygonization and is able to provide us with user-defined resolution of triangles.
|
