Volume Rendering Research

My dissertation research deals with the problem of speeding up volume rendering of univariate, regularly spaced data, such as medical data. Recent advances in software algorithms have produced software which is extremely efficient at sampling voxels. Recent advances in hardware have produced implementations which are extremely efficient at interpolating samples.

In order to surpass the rendering speed of these kinds of techniques, it will be necessary to sample far fewer voxels than are required to construct the final image. To do this, we will preprocess the volume to find homogeneous regions in order to reduce the number of samples needed for rendering an image with a controlled amount of error. My research deals with the problems of:

• Finding fast preprocessing techniques to compute homogenous regions
• Apply this preprocessed volume to the already accelerated algorithms in the following rendering domains:
  • Ray-casting
  • Object-order traversal

Pre-process volume data

• Identify region boundaries.
  
• Use 3-D Distance Transform to compute region sizes
  
• Extract the largest, non-overlapping homogeneous regions
  

Accelerated Scanning Algorithm

• Do fast 2-D blit of 3-D extracted homogeneous regions
  
• Use Lacroute and Levoy [SIGGRAPH 1994] to scan all non-empty, non-occluded, non homogeneous regions
  
• Blend of above images
  
• Is close to a clean rendering, but renders twice as fast with < 4% image error
  

• My dissertation, June, 1997, University of Alabama at Birmingham
• Proceedings of Visualization '97