GPU-BASED VOLUME RAY-CASTING SUPPORTING

SPECULAR REFLECTION AND REFRACTION

Timo Ropinski, Klaus Hinrichs

Visualization and Computer Graphics Research Group (VisCG), University of M

unster, Germany

Jens Kasten

Konrad-Zuse-Zentrum f

ur Informationstechnik, Berlin, Germany

Keywords:

Volume rendering, GPU-based ray-casting, Advanced illumination.

Abstract:

Nowadays mostly local illumination models are used when rendering volumetric data. When computing global

light effects, interactive frame rates are usually hard to achieve. We present an extension of GPU-based volume

ray-casting, which allows to compute specular reﬂection and refraction effects at interactive frame rates on

current commodity graphics hardware. In contrast to other techniques proposed for integrating these effects

into volume rendering, our technique does not constrain the type of rendering used, i. e., it can be used with

DVR as well as isosurface rendering.

1 INTRODUCTION

Much research has been conducted in the past to

achieve interactive frame rates for volume rendering

on consumer graphics hardware. For example, with

GPU-based volume ray-casting (Roettger et al., 2003)

interactive frame rates are possible while generating

a high-quality rendering. Due to these performance

aspects it becomes possible to integrate more sophis-

ticated illumination models to increase the visual re-

alism of volume rendered images (see Figure 1).

The proposed technique modiﬁes GPU-based ray-

casting by processing a ray-caster multiple times with

different entry and exit points. Thus, we are able to

use arbitrary ray-caster modules, potentially support-

ing different rendering styles, by just transforming

their input points. Since the proposed implementa-

tion exploits the capabilities of current graphics hard-

ware and achieves interactive frame rates while sup-

porting global illumination phenomena, we support

full interactivity. Thus, the transfer function can be

changed interactively, and it is possible to deﬁne dif-

ferent materials, e. g., more reﬂective or glassy ones.

One important aspect is our progressive reﬁnement

of the resulting rendering. By using this reﬁnement,

it becomes possible to support specular reﬂection as

well as refraction even on older graphics hardware by

still allowing interactive exploration.

2 RELATED WORK

A lot of research has been conducted with the goal

to allow interactive frame rates when ray-tracing vol-

umetric data sets. Kajiya and von Herzen (Kajiya

and Herzen, 1984) propose a volumetric ray-tracing

system, which allows to simulate scattering besides

the typical ray-tracing effects like specular reﬂection.

The ray-tracing technique presented by Marmitt and

Slusallek is more interactive, but constrained to iso-

surfaces (Marmitt and Slusallek, 2006). Since inter-

activity is important to be able to modify important

rendering parameters, e. g., the thresholding or the

transfer function, Marmitt et al. review different ap-

proaches for interactive ray-tracing of volumetric data

in a state-of-the-art report (Marmitt et al., 2006).

Besides ray-tracing, various publications also con-

sider refraction in volume graphics. Rodgman and

Chen describe different approaches, which exploit

a ray tracer to ﬁnd refractive indices of materi-

als (Rodgman and Chen, 2001). Li and M

uller aim

at smooth gradients by proposing a B-spline kernel

for gradient ﬁltering (Li and Mueller, 2005).

One approach to integrate specular reﬂection and

refraction into a GPU-based volume ray-caster has

been presented by Stegmaier et al. (Stegmaier et al.,

2005). They describe a ray-casting framework for

generating highly appealing renderings which incor-

219

Ropinski T., Hinrichs K. and Kasten J. (2009).

GPU-BASED VOLUME RAY-CASTING SUPPORTING SPECULAR REFLECTION AND REFRACTION.

In Proceedings of the Fourth International Conference on Computer Graphics Theory and Applications, pages 219-222

DOI: 10.5220/0001784402190222

 SciTePress

(a) specular reﬂections

(b) specular reﬂections and shadows

Figure 1: Interactive volume renderings with a higher de-

gree of realism can be generated by exploiting specular re-

ﬂection effects. A CT scan of a hand rendered with specular

reﬂections (a). By also adding shadows the degree of real-

ism can be further increased (b).

porate these effects. While they are more focussing on

the exchangeability of different ray-casters, we aim at

the combination of specular reﬂection and refraction

effects within a single image.

3 EXTENDING GPU-BASED

RAY-CASTING

3.1 Higher Order Entry and Exit Points

To integrate global illumination effects into a GPU-

based volume ray-caster, we not only cast one ray per

pixel, but also trace rays of higher order, i. e., we ex-

tend the ray-caster to become a ray-tracer. To be able

to trace these rays, we need higher order entry and

exit points for them.

The workﬂow of our ray traversal approach can

be subdivided into four subsequent stages as shown in

Figure 2. In the ﬁrst stage the initial unmodiﬁed entry

and exit points are computed as described in the pre-

vious subsection. By using these points the ﬁrst order

rays can be cast to generate one intermediate image

as well as the computed ﬁrst hit points, i. e., the posi-

tions in volume space where a ray ﬁrst encounters a

visible medium. The intermediate image contains the

shading result achieved for each pixel, when casting

a ray from the given entry point to the ﬁrst hit point.

While the intermediate image is cached in order to be

able to blend it in the last stage, the ﬁrst hit points are

used as the entry points in the next recursion step. To

increase the foot print of the voxels encountered at the

ﬁrst hit point, we alter the ﬁrst hit point computation

by sampling one step further into the volume. This

ensures that the encountered medium is sufﬁciently

penetrated and thus more clearly visible in the inter-

mediate image. The exit points for the next recur-

sion step are computed in the second stage, as shown

in Figure 2. Based on the entry position p and the

direction d of a higher order ray r, we compute the

intersection between r and the bounding box of the

volume. After this computation has been performed,

we can hand the entry and exit points to the subse-

quent ray-caster, which performs the rendering in the

third stage. When all ray-casters have ﬁnished ren-

dering, the ﬁnal image can be computed by blending

all available intermediate shading images within the

fourth stage.

To compute the ray direction d for rays reﬂected

on a specular surface, we consider the normal of this

surface. Thus, similar to the computation of the spec-

ular reﬂection in the Phong illumination model, the

outgoing ray can be computed by considering the in-

coming ray and the current surface normal.

In cases where a refraction occurs, the incoming

ray is bent at the surface based on Snell’s law. To

compute the bending angle the refraction indices of

the adjacent media, for which the refraction should

be computed, has to be known. When assuming that

these indices are n

for the medium which is left by

the ray, and n

for the medium which is entered by the

ray, we can compute the bending angle θ based on the

incoming angle φ as follows:

cos(θ) =

1 −





· (1 − (cos(φ))

Thus, we can compute the direction of the leaving

ray A

vec

as follows:

vec





vec



|cos(φ)| − cos(θ)



vec

where E

vec

is a vector representing the incoming

ray and N

vec

is the normalized gradient representing

the current surface normal.

GRAPP 2009 - International Conference on Computer Graphics Theory and Applications

220

existing

entry exit points

blending

with transfer function

entry exit point generation

for higher order rays

ray-casting

final rendering

blended intermediate

images

specularity mapping function

intermediate

image

first hit

points

Figure 2: The workﬂow of the proposed technique can be subdivided into four subsequent stages. Initially the unmodiﬁed

entry and exit points are computed for the ﬁrst order rays (stage 1). Afterwards, for each recursive step, the exit points are

calculated based on the ﬁrst hit points of the previous step (stage 2). A ray-caster computes intermediate images containing

the shading results for a ray segment, based on the generated entry and exit points (stage 3). Finally all intermediate images

are blended into the ﬁnal image using a specularity mapping function (stage 4).

In general we have to consider total reﬂection in

cases the incoming ray hits the object under a very

ﬂat angle, i. e., no refraction but specular reﬂection

occurs. The critical angle for which total reﬂection

occurs can be computed as follows:

crit

= sin

−1

(

Thus, when the incident angle φ exceeds φ

crit

, we

compute a specular reﬂection ray instead of a refrac-

tive one.

4 RESULTS

Table 1 shows the frame rates for different recursion

depths and a ﬁxed sampling rate, which is twice the

maximum grid resolution of the rendered data set. To

show the scalability, we have tried several data sets

having different resolutions and have also altered the

screen resolution. While the recursion depth has an

inﬂuence on the ray length to be traversed, the latter

inﬂuences the screen resolution of rays to be traced.

As can be seen from the table, we still achieve inter-

active frame rates for moderately sized data sets when

using a screen resolution of 512 × 512 pixels.

Parts of the hand shown in Figure 3 are rendered

semi-transparently, Figure 3 (left) shows a rendering

without, Figure 3 (right) with refraction. As it can be

seen, refraction gives the semi-transparent medium a

more glassy effect.

Table 1: The average frame rates achieved for different re-

cursion depths with precomputed gradients, as measured on

a GeForce 8800GTX graphics board. We have compared

different data set sizes and screen resolutions.

data set recursion screen resolution

depth 256

512

Hand 0 57 fps 39 fps

256 × 256 × 147 1 38 fps 22 fps

2 30 fps 14 fps

3 24 fps 12 fps

Teapot 0 55 fps 38 fps

256 × 256 × 178 1 40 fps 25 fps

2 33 fps 19 fps

3 30 fps 17 fps

Cornell Box 0 59 fps 47 fps

256 × 256 × 256 1 38 fps 19 fps

2 31 fps 12 fps

3 18 fps 10 fps

The proposed technique can also be combined

with other advanced illumination techniques devel-

oped for volume rendering. Figure 4 shows the com-

bination with ambient occlusion and deep shadow

maps (Hadwiger et al., 2006). By exploiting these

illumination techniques, a high level of realism can

be achieved, while still maintaining interactive frame

rates.

GPU-BASED VOLUME RAY-CASTING SUPPORTING SPECULAR REFLECTION AND REFRACTION

221

(a) no refraction (b) refraction

Figure 3: The hand data set rendered without refraction (a), and by applying refraction (b).

Figure 4: A synthetic Cornell box data set rendered with

specular reﬂections, shadows and ambient occlusion.

5 CONCLUSIONS

We have proposed how to extend an interactive GPU-

based volume ray-caster to support specular reﬂection

and refraction effects. By exploiting the capabilities

of current graphics hardware, our technique can be

applied by still achieving interactive frame rates. The

presented approach can be easily integrated into ex-

isting volume rendering frameworks, which are based

on GPU-based volume ray-casting.

ACKNOWLEDGEMENTS

This work was partly supported by grants from the

Deutsche Forschungsgemeinschaft (DFG), SFB 656

MoBil M

unster, Germany (projects Z1, PM5). The

presented concepts have been integrated into the

Voreen volume rendering engine www.voreen.org.

REFERENCES

Hadwiger, M., Kratz, A., Sigg, C., and B

uhler, K. (2006).

Gpu-accelerated deep shadow maps for direct vol-

ume rendering. In GH ’06: Proceedings of the

21st ACM SIGGRAPH/Eurographics symposium on

Graphics hardware, pages 49–52, New York, NY,

USA. ACM Press.

Kajiya, J. T. and Herzen, B. P. V. (1984). Ray tracing vol-

ume densities. In SIGGRAPH ’84: ACM SIGGRAPH

1984 Papers, pages 165–174. ACM Press.

Li, S. and Mueller, K. (2005). Accelerated, high-quality re-

fraction computations for volume graphics. In Inter-

national Workshop on Volume Graphics 2005, pages

73–229.

Marmitt, G., Friedrich, H., and Slusallek, P. (2006). Inter-

active Volume Rendering with Ray Tracing. In Euro-

graphics State of the Art Reports.

Marmitt, G. and Slusallek, P. (2006). Fast Ray Traversal

of Tetrahedral and Hexahedral Meshes for Direct Vol-

ume Rendering. In Ertl, T., Joy, K. J., and Santos,

B., editors, Proceedings of Eurographics/IEEE-VGTC

Symposium on Visualization (EuroVIS) 2006, pages

235–242, Lisbon, Portugal. n/a.

Rodgman, D. and Chen, M. (2001). Refraction in dis-

crete raytracing. In International Workshop on Volume

Graphics 2001.

Roettger, S., Guthe, S., Weiskopf, D., Ertl, T., and Strasser,

W. (2003). Smart hardware-accelerated volume ren-

dering. In VISSYM ’03: Proceedings of the Sympo-

sium on Data Visualisation 2003, pages 231–238. Eu-

rographics Association.

Stegmaier, S., Strengert, M., Klein, T., and Ertl, T. (2005).

A simple and ﬂexible volume rendering framework

for graphics-hardware–based raycasting. In Proceed-

ings of the International Workshop on Volume Graph-

ics ’05, pages 187–195.

GRAPP 2009 - International Conference on Computer Graphics Theory and Applications

222