Research


Visualization and Computer Graphics on Isotropically Emissive Volumetric Displays


Our TVCG paper discusses the pros and cons of emissive 3D displays like the perspecta display. This research explains and gives the right rendering model that must be followed to display accurate graphics on those devices. While it is demonstrated that no perfect rendering is possible with such technology, the paper still show that realistic graphics and visualization is still possible.

More can be found about this work on the physics.org website, or similar websites (1,2).

 



 


General Research on Ray Tracing


Some work have been done on extending algorithms for Ray-Tracing, and studying complexity of some algorithms. Restricted BSP trees have been demonstrated as an extension to regular Kd-Trees. Later on, we demonstrated row-tracing, a mix between ray-tracing and rasterization, a new technique combining advantages from both rasterization and ray-tracing.  

Research and knowledge of Ray-Tracing has been considerably extended further with the support of EPSRC. This innovative work will be presented in due time.

 


Low Complexity Maximum Intensity Projection


A new way to perform MIP renderings at 2-3 FPS, on a 512^3*12b dataset, with any Tranfer Function (LUT) and on current low-cost CPUs...

We first demonstrate the fundamental result that MIP can be achieved with a O(n2) average complexity. We also propose an implementation satisfying our theoretical complexity analysis. Apart from the theoretical importance of this result, the actual implementation is very efficient as well. The image here below has been computed from a LUT using the full intensity range, and with a sampling distance of 0.5. The rendering rate is greater than 1 fps on a 1.4 GHz Athlon CPU.  

 


Xmastree1 (512^2*499) rendered at ~1 FPS


 

1. The CT-dataset XMasTree was generated from a real world Christmas Tree by the Department of Radiology, University of Vienna and the Institute of Computer Graphics and Algorithms, Vienna University of Technology.


Instant Volumetric Understanding

In this work, we tried to propose better, but simple, rendering models for fast perception and understanding of volumetric datasets. Also, we tried to show why optical models are often not suitable to volume rendering, especially when a fast insight of the volume is required.   
 
 
 
 
 
 
 


Fast Ray-Casting


Due to the limitation of our previous technique (Parametric Isosurfaces) on the visualization of isosurfaces, we have then focused on a new way to compute ray-casting called object-order ray-casting (published in IEEE Visualization 2002). Colors and opacities are evenly sampled along the rays, but the algorithm works now in an object-order way. This way, we manage to combine the advantages of ray-casting with trilinear interpolation and the efficiency of object-order methods. We have also developed a new hidden volume removal technique that compensates the loss of early ray termination. The resulting algorithm is probably one of the best volume rendering algorithms so far. It is as fast as shear-warp and provides a very good accuracy.
 
 

Click the images for high resolution video (AVI intel indeo 5.1). Skull 6.5 MB, Engine 16.5 MB.


Parametric Isosurfaces

In EUROGRAPHICS 2001 and SMI 2002, we showed how to interactively visualize a high-quality quadratic isosurface, which has a lot of advantages when compared to other interpolation filters like the trilinear or gaussian filters. This technique is well-suited for the interactive visualization and modelisation of implicit surfaces because no intermediate triangular mesh is needed.

Click the image for video (AVI intel indeo 5.1). 4.7MB.


Quantized Voxels

Quantized voxels (Volviz 2000) can be used inside as an efficient way to project voxels. Our first volume rendering algorithm was able to project (and also uniformly classify and shade) up to 4.5 millions cells per second or 1 million Gaussian splats on a Pentium II 450 Xeon. Those results were feasible with the use of an orthogonal projection. Some other publications on volume rendering use this principle.