Spline-Based Gradient Filters for High-Quality Refraction
Computations in Discrete Datasets
Eurographics / IEEE VGTC Symposium on Visualization 2005
Stony Brook University
Abstract:
Based on the finding that refraction imposes significantly higher demands onto gradient filters than illumination and shading, we evaluate the family of spline filters as a good alternative to the cubic filters, which so far have served as the gold standard of efficient yet high-quality interpolation filters in present visualization applications. Using a regular background texture to visualize the refractive properties of the volumetric object, we also describe an efficient scheme to achieve the effects of supersampling without incurring any extra raycasting overhead. Our results indicate that splines can be superior to the Catmull-Rom filter, with potentially less computational overhead, also offering a convenient means to adjust the extent of lowpassing and smoothing.
Full Paper (PDF file, 400 KB)
Presentation (PPT file, 2.4 MB)
B-Spline frequency properties:
|
|
|
|
|
Figure: Frequency response of (from left to right) (1) B-spline (2) B-spline 3 and smooth B-spline 3 (3) derivative B-spline |
Volume rendering results:
 |
 |
 |
 |
| Catmull-Rom Cubic | B-spline 2 | B-spline 3 | B-spline 6 |
|
Figure: Rendering results of Marschner-Lobb. |
|
|
|
|
|
|
Analytical |
Central Difference |
Linear |
Catmull-Rom Cubic |
|
|
|
|
|
|
B-spline 3 |
B-Spline 3 w/o prefiltering |
B-spline 3, smooth λ=0.1 |
B-spline 3, smooth λ=1 |
|
Figure: Refracted result of sphere with different filters and without supersampling. |
| (a) Traditional Supersampling |
|
|
|
| (b) Post-Refraction Supersampling |
|
|
|
|
Central Difference |
Catmull-Rom Cubic |
B-Spline 3, smooth λ=1 |
|
Figure: Refracted result of sphere with different supersampling methods: traditional supersampling v.s. post-refraction supersampling. |
lshengyi at cs.sunysb.edu, Computer Science Department, Stony Brook University, NY 11794.
