Preprocessing

Given a set of visualization parameters, the goal of the preprocessing step is to classify voxels of the volume into voxels which possibly contribute to an image and voxels which do not contribute to an image. The classification criteria depend on the chosen opacity transfer function, the desired compositing method, and the degree of freedom which should be provided for further manipulation of the transfer function. Generally speaking, the more voxels are classified as irrelevant, the fewer data has to be processed during projection, and the faster the rendering becomes.

Different rendering methods and data characteristics require different extraction strategies:

• Object surfaces (iso-surfaces): To obtain a representation for the iso-surface, the volume is scanned for transitions between voxels within and outside an object. If a voxel is located within the object (data value greater or equal than the threshold for an iso-surface) and if it has at least one 26-connected neighbor outside the object, it belongs to the object's surface and is considered to be relevant. The relevant voxels identified during the scan correspond to a 6-connected surface within the volume at the specified threshold value or object boundary. The 6-connectedness of the surface voxels is required to ensure that no holes appear during (shear/warp-based) rendering due to displacement of voxels within successive slices. Just a few percent of all voxels within a volume usually belong to a meaningful object surface.
• Maximum Intensity Projection (MIP): Most approaches to optimize the performance of MIP rendering aim at excluding voxels from the traversal and rendering process, which contain less-important information like low-valued background noise. In fact, in addition to this low-importance data, there is usually a remarkable amount of voxels which never contribute to a MIP image. A voxel $V$ does never contribute to a MIP (and can be discarded from rendering) if all possible viewing rays through the voxel hit another voxel $W$ with $d(W)\geq d(V)$, either before or after passing through $V$, where $d(V)$ is the data value at voxel $V$. This fact can be exploited when original voxel values are used for rendering using nearest neighbor interpolation, as it is done within the presented approach. Several elimination strategies can be applied to identify such voxels, ranging from examining just the direct neighbors of each voxel, to tracking of voxel influences into distant parts of the volume. By also considering viewing direction into the elimination process, up to 75% of all voxels can be discarded.
• Gradient Magnitude Modulation: If a transfer function is used which modulates voxel opacity according to gradient magnitude, gradient magnitude can be used as an elimination criterion.

mailto:mroz@cg.tuwien.ac.at.