Visualization of volumetric datasets (Q2781531)

Volume rendering represents a technique for visualizing 3D arrays of sampled data. This technique has many applications in scientific visualizations and medical image manipulation, but remains a computational expensive technique. This thesis presents a fast rendering technique for volumetric datasets used in a user friendly environment that allows an interactive access to the different rendering parameters. Using this facility, the volumetric data like a textured cube is drawn. The texture is created by integrating along rays casted into the volume. By integration along the ray the volume data is projected onto the boundary of the volumetric data. The decomposition of the problem into different modules and implementation details are presented as well. NEWLINENEWLINENEWLINEHybrid visualization of volumes and different surface types is considered as well. As data representation a rectilinear, uniformly spaced grid is used. This grid structure together with an object ordered ray casting method permits to take benefit of acceleration techniques like ray-path template and spatial coherence. Object ordered ray casting uses the resolution of the casted object ensuring the consistence of the ray-path template from one ray to another. The ray path template saves the offsets of the voxels intersected by the ray. Using this offset one knows at each step along the ray which is the next voxel to process. The voxel's offset from the current position is stored in the ray-path template. Using a scanline ordered ray-casting on a rectilinear uniform grid, one can take benefit of spatial coherence in order to skip connected regions of empty voxels. By changing the order of integration from ray ordered to scanline ordered one can take benefit of empty voxels coherence along the scanlines.NEWLINENEWLINENEWLINEUsing this approach on a highly coherent empty voxel structure one can induce a speedup of 5 to 10 times from the brute force approach. By using both acceleration techniques combined with early ray termination the author achieves a rendering of 1-3 seconds for a \(256^3\) voxels volumetric dataset. Some examples of the most important scientific fields in which the presented volume rendering techniques and their enhancements play an important role are shown in order to analyse and visualize real and numerically simulated experiments.