The main extensions added to Levoy's Volume Rendering presentation of 1988, was to integrate multiple surfaces classification scheme rendered with different colours. Also, in the original paper, no algorithms were introduced to render the volume from different views, only views parallel to the volume axes were generated. A method based on direction cosines, implementing rotation around one of the volume axes, was included in this project.
However, there are other methods to project a volume to an image along any arbitrary view direction. This project focused on ray-casting algorithms, but another direction could have been to consider voxel projection methods. Voxel projection is the process of traversing the volume data set and projecting each voxel onto the image plane. Splatting (Wes90) is one such voxel projection method. It is also called footprint since a single voxel forms a projection in the image plane that spreads over many pixels. Voxel projection rendering methods are often considered because they are easier to parallelize than ray-casting. Volume Rendering, especially, has a need to use hardware effectively, because they are computationally intensive due to size of the data sets. In Volume Rendering efficiency is sometimes demanded because interactivity is important to visualization. Voxel projection could be implemented as further work, in order to improve the running of the application.
Nevertheless, the results obtained by Levoy Volume Rendering methods was successful. The different tools that implement classification, compositing and projection work well. However, it has been discovered that Levoy's classification works poorly for certain data sets. The data set needs to be pre-filtered and the surfaces, which are desired to be rendered, need to have a certain kind of properties. For example, it is hardly possible to render the brain surface from a CT-scan because, the surface of brain tissue is too convoluted and do not form a comprehensible image. Furthermore, the density of the interior of the brain is the same as the density of the cerebra-spinal fluid. Levoy's classification method works best for non-interpenetrating layers of surfaces of different densities (like an onion with each layers having different densities).
Volume rendering is tedious and complex because it is necessary to get a score of details exactly right to get a successful volume visualization. Volume rendering also demand a lot of experimentation to set the parameter appropriately. However, when it works, volume rendering produces output images that are satisfying and rewarding.