4.3 Data preparation

In order to extract sensible images from the volume data, pre-filterings need to be done on the initial data. Regions of similar density need to be averaged and smoothed out, and edges need to be reinforced. Also overall noise needs to be removed. For noise reduction, a gaussian blur filter can be applied and for boundary re-enhancement a gradient filter is used. Pre-filtering high resolution and noisy data is a field of research by itself, and will not be considered in this project.

The three dimensional data recorded by scanners is not usually uniformly sampled. The recorded rate can differ in the three dimensions, i.e. the interval between the points is often not the same in all three directions. Two approaches are possible : either to initially resample to create isometric voxels, or use non-uniform interpolation during projection.

Segmentation is a preprocessing step performed before the actual rendering computations. It separates the data into structural units by putting different labels on different categories of voxels. For example, in a medical image, segmentation will mark the different type of voxels according to which anatomical component they belong, e.g. skin, bone, muscle. This step by itself is also another topic, it can involve user-defined segmentation or semi-automatic one. Levoy's volume rendering implementation does not require explicit segmentation, which is handled by the ulterior classification step that uses thresholding.