by Ivan Viola, Armin Kanitsar, Meister Eduard Gröller

Abstract

Frequency domain volume rendering (FVR) is a volume rendering technique with lower computational complexity as compared to other techniques. In this paper the FVR algorithm is accelerated by factor of 17 by mapping the rendering stage to the GPU. The overall hardware-accelerated pipeline is discussed and the changes according to previous work are pointed out. The three-dimensional transformation into frequency domain is done in a pre-processing step. The rendering step is computed completely on the GPU. First the projection slice is extracted. Four different interpolation schemes are used for resampling the slice from the data represented by a 3D texture. The extracted slice is transformed back into the spatial domain using the inverse Fast Fourier or Fast Hartley Transform. The rendering stage is implemented through shader programs running on programmable graphics hardware achieving highly interactive framerates.

Keywords:

Download full paper

Ivan Viola, Armin Kanitsar, Meister Eduard Gröller, "GPU-based Frequency Domain Volume Rendering", in proceedings of SCCG 2004, pages 49-58, second-best paper award! Download: hwfvr.pdf (1,197KB).

Figures in the paper

Figure 1:

Frequency domain volume rendering pipeline. Instead of projecting the data in the spatial domain, slicing in frequency domain is performed after an off-line pre-processing step.
Figure 2:

Setup for the resampling part shows relationship between 3D texture, proxy slice geometry and rendering target.
Figure 3:

Blending factor estimation. The resolution of the original texture is 8 x 8. The texture coordinates are multiplied with the original resolution. The blending factors are equal to the fractional parts of these new coordinates.
Figure 4:

Filter kernel of width four and corresponding 1D floating-point RGBA texture storing the discretized kernel. Each channel stores one kernel tile. A single texture fetch at position X returns 4 weight values.
Figure 5:

State diagram of GPU-based FFT.
Figure 9:

Quality comparison using the bonsai tree dataset. Projection slice resolution is 256 x 256 and 512 x 512. Low projection slice resolution results into image distorted by overlapping copies. This effect is removed when using a sufficient resampling resolution.
Figure 11:

Result of rendering the X-mas tree test data set. The images show the rendering quality according to the used interpolation scheme. Nearest neighbor interpolation exhibits noticeable artifacts, which are eliminated by tri-linear interpolation, respectively by higher-order interpolation schemes like tri-cubic interpolation using cubic B-spline filter or windowed sinc filter using a Blackman window of width four.
Figure 12:

Other datasets rendered with projection slice resolution 512 x 512 using tri-cubic B-spline reconstruction filter.

BibTeX Entry

@InProceedings{sccg-viola04,
  author = {Ivan Viola and Armin Kanitsar and Meister Eduard Gr{\"o}ller},
  title = {GPU-based Frequency Domain Volume Rendering},
  booktitle = {Proceedings of SCCG'04},
  year = {2004},
  pages = {49--58},
}