A GENERALIZATION APPROACH

FOR 3D VIEWING DEFORMATIONS

OF SINGLE-CENTER PROJECTIONS

Matthias Trapp and J

urgen D

ollner

Hasso-Plattner-Institut, University of Potsdam, Germany

Keywords:

Real-time Panorama, Non-Planar Projection, Fish-Eye Views, Projection Tiles.

Abstract:

This paper presents a novel image-based approach to efﬁciently generate real-time non-planar projections of

arbitrary 3D scenes such as panorama and ﬁsh-eye views. The real-time creation of such projections has a

multitude of applications, e.g., in geovirtual environments and in augmented reality. Our rendering technique

is based on dynamically created cube map textures in combination with shader programs that calculate the

speciﬁc projections. We discuss two different approaches to create such cubemaps and introduce possible

optimizations. Our technique can be applied within a single rendering pass, is easy to implement, and exploits

the capability of modern programmable graphics hardware completely. Further, we present an approach to

customize and combine different planar as well as non-planar projections. We have integrated our technique

into an existing real-time rendering framework and demonstrate its performance on large scale datasets such

as virtual 3D city and terrain models.

1 INTRODUCTION

This work introduces a concept to compensate for the

ﬁeld-of-view (FOV) limitations of the classical pin-

hole camera rendering pipeline. It has been developed

to enable the application of non-planar projection on

standard consumer graphics hardware in real-time.

Examples are omni-directional panorama for non-

planar screens or spherical dome projections (Bourke,

2004). This work focuses on real-time modiﬁcations

of perspective views that are possible due to the recent

hardware developments (Blythe, 2006).

The research in the ﬁeld of non-planar or non-linear

projection distinguishes mainly between two dif-

ferent projection types: projections with a single

projection center (SCOP) (Carlbom and Paciorek,

1978) or with multiple projection centers (MCOP)

(Rademacher and Bishop, 1998). Our approach is

limited to SCOP. In spite of non-planar projections

screens (Nelson, 1983), this technique can also be

used for rendering effects in games as well as to

improve visibility if used in virtual landscapes (Rase,

1997) or city environments. The user can beneﬁt

from extreme perspectives (Glaeserm, 1999) and

large FOV angles by having a reasonable survey

of the scene (Glaeser and Gr

oller, 1999) as well

as a better size and depth perception (Polack-Wahl

et al., 1997). Our method exploits the technique of

dynamic environment mapping (in terms of cube map

texturing) in combination with the programmable

GPU. The separation of projection calculation and

cube map texture creation enables a broad range of

optimization techniques.

Figure 1: A cylindrical projection with a horizontal FOV of

270

◦

and a vertical FOV of 90

◦

Existing image based approaches for non-planar pro-

jections suffer mainly from the lack of interactive ca-

pabilities if used with complex geometric scenes such

as virtual 3D city models or for large viewports. This

can be explained by the trade-off between the gener-

ality of the proposed frameworks and their efﬁcient

reproduction. Furthermore, the parameterizations are

complex and cannot be controlled by the user intu-

itively (Brosz et al., 2007).

Our approach was inspired by (van Oortmerssen,

163

Trapp M. and Döllner J. (2008).

A GENERALIZATION APPROACH FOR 3D VIEWING DEFORMATIONS OF SINGLE-CENTER PROJECTIONS.

In Proceedings of the Third International Conference on Computer Graphics Theory and Applications, pages 163-170

DOI: 10.5220/0001098201630170

 SciTePress

2002). This CPU-based technique renders six views

of FOV 90 in each direction. Afterwards, a table is

used to transform these pixels to one single view ac-

cording to ﬁsheye and panorama projections.

Our main contribution consists of a simple parame-

terizable approach to combine planar as well as non-

planar projections seamlessly via so-called projection

tiles. Therefore we introduce an efﬁcient image-based

creation method for non-planar projections that can

be applied in a single rendering pass. Our rendering

technique fully exploits current programmable con-

sumer graphics hardware. The presented concept is

easy to implement into existing rendering real-time

frameworks and can be combined with other tech-

niques that modify the image synthesis.

The paper is structured in the following way: Section

2 describes the related work. Chapter 3 describes the

basic concept of our approach whilst Section 4 intro-

duces a novel generalization schema for non-planar

projections. Section 5 explains the implementation

details. Section 6 presents possible results and appli-

cations as well as discusses the performance and lim-

itations of our rendering technique. Section 7 shows

ideas for future work and Section 8 draws some con-

clusions.

2 RELATED WORK

This section gives an overview of research in the ﬁelds

of SCOP and MCOP projections. There is a vast

amount of literature covering foundations and appli-

cations of non-planar as well as non-linear projec-

tions. In (Brosz et al., 2007) a sophisticated overview

is presented.

2.1 SCOP Approaches

To achieve distortions or special projections of the

3D scene, the pinhole camera is extended in several

ways. In (Bayarri, 1995) a procedure is proposed that

is based on the computation of new absolute coordi-

nates to be transformed through an adaptive projec-

tion matrix. A ﬂexible adaptive projection framework

is described by Brosz et.al. (Brosz et al., 2007) that

enables the modeling the modeling of linear, non-

linear, and hand-tailored artistic projections. It uses

ray-casting and scanline rendering algorithm where

polygonal coordinates are changed by a vertex shader.

The generality of the framework makes efﬁcient pro-

jection difﬁculty, especially for large scale scenes.

Distortions as sub-category of geometric registration

or image warping are discussed in (Gustafsson, 1993)

and (Glasbey and Mardia, 1989). A warping func-

tion is applied to each pixel to determine its new color

value.

An image stitching approach for panorama image

generation can be found in (Szeliski and Shum,

1997). In (Turkowski, 1999) a method is demon-

strated to generate environment maps from ﬁsheye

photographs. Besides the issues of nonlinear perspec-

tive deformation described in (Yang et al., 2005; H.

et al., 1999; Yang et al., 2003; Bourke, 2000; Swami-

nathan et al., 2003) we ﬁnd also lens taxonomies

(Neumann and Carpendale, 2003; Leung and Apper-

ley, 1994). These approaches use a regular mesh tex-

tured with a 2D texture that contains the rendered

scene or an image. The displacement of the mesh ver-

tices together with the texture mapping process gener-

ates the particular distortion effect. These approaches

are limited regarding the FOV which can be achieved.

Carpendale researched the usage of image defor-

mation in the context of information visualization

(Carpendale and Montagnese, 2001). The application

of ﬁsheye views in information visualization is dis-

cussed in (Rase, 1997). Applications for view distor-

tions in ray-tracing software are described in (Gr

oller

and Acquisto, 1993; Coleman and Singh, 2004).

2.2 MCOP Approaches

Additional to SCOP imaging there are non-planar

projection surfaces that require multiple perspectives,

e.g., a number of images from different center of pro-

jections (Wood et al., 1997). The goal is to keep

qualities of global scene coherence, local distortions,

and shading results from the changes in perspective

of each projection. View-independent rendering, ex-

pressive CG imagery or animation are just a few ap-

plications for this area of scientiﬁc research.

1 Apart from slit cameras, (Glassner, 2000) intro-

duced the Cubist Camera that presents many interpre-

tations and points of view simultaneously. The tech-

nique uses nonlinear ray tracing that handles lighting

but can cause artifacts.

Given a scene geometry and an UI to position local

and master cameras (Agrawala et al., 2000), Artis-

tic Multiprojection Rendering provides a tool for cre-

ating multi-projection images and animations. The

principal item is the algorithm that solves occlusions

of the scene objects, each rendered by a local cam-

era. This is able to handle ﬁxed distortions and cre-

ates surrealistic, toony styles but cannot solve lighting

and shadow problems.

A fresh Perspective (Singh, 2002) introduces also an

interactive approach that does neither handle illumi-

nation issues nor control global scene coherence. The

difference: the resulting nonlinear perspective image

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

164

ABC

Figure 2: Comparison between a classical perspective projection with a FOV of 45

◦

(A) and a spherical projection with a

FOV of 260

◦

(B). Sub-ﬁgure (C) shows the same projection with an off-axis vector O = (0.8,0,0).

of an object is potentially inﬂuenced by all cameras.

Constitutive on that is RYAN (Coleman and Singh,

2004). This interactive system integrates into the con-

ventional animation work ﬂow. It distorts the scene

geometry prior to the linear perspective transforma-

tion. The result will appears nonlinearly projected.

Just like (Agrawala et al., 2000), it uses two kinds

of cameras: boss (traditional linear perspective) and

lackey (represent local linear views). The illumina-

tion is done by blending illumination of boss and

lackey cameras or setting a single view point for light-

ing.

3 BASIC CONCEPT

The concept of our approach is based on two com-

ponents: a dynamic environment map (Heidrich

and Seidel, 1998) and fragment shader functionality

(NVIDIA, 2005). First, the virtual environment is

rendered into a cube map texture. The cube map tex-

ture is a fully hardware accelerated feature (NVIDIA,

2004) and can be constructed by using single or multi-

pass rendering (see Section 5.1).

To derive a non-planar projection, we have to deter-

mine a cube map sampling vector S = (x,y,z) ∈ D

for each fragment F

= (s,t) ∈ D

. Formally, we de-

ﬁne a projection function δ

) = S for a projection

P as:

: D

−→ D

(s,t) 7−→ (x,y,z)

Where D = [−1; 1] ⊂ R is a normalized coordinate

space. For example, a horizontal cylindrical projec-

tion C can be formulated as instance δ

,α, β) = S

with an horizontal FOV of 2 ·α and a vertical FOV of

2 ·β :

x = cos(s ·α) y = t ·tan(β) z = sin(s ·α)

Further, a spherical projection S with an FOV of γ can

be expressed as δ

,γ) = S with:

x = sin(θ) ·cos(φ)

y = sin(θ) ·sin(φ)

z = cos(θ)

φ = arctan(t, s)

θ =

√

·γ/2

This procedure assumes an user orientation towards

the negative z-axis. If the cube map texture is created

in the standard orientation (Figure 3.A), we would

have to correct S by transforming it with respect to

the current camera parameters. The transformation

matrix C is an orthonormal base constructed from the

current look-to vector L

= (x

) ∈D

, look-up

vector L

= (x

) ∈ D

and the cross product

= (x

) = L

×L

so that:

C =









⇐⇒ L

•L

6= 0

Following this, the ﬁnal sampling vector V for a pro-

jection P and a fragment F

is calculated via:

V =



C ·δ

·s)



−O

The vector O ∈ D

is denoted as off-axis vector

(Bourke, 2004). Figure 2.C demonstrates an exam-

ple for an off-axis projection. The scalar term s ∈ D

can be interpreted as a zooming parameter.

To avoid unnecessary calculations at runtime, the dis-

tortion vector V can be stored into a high-precision

ﬂoating-point normal map (Kilgard, 2004). This re-

duces the shader execution costs for projection calcu-

lation to two texture look-up operations and enables a

new technique for combining different projection us-

ing projection tiles.

4 PROJECTION TILES

This method enables the combination of planar pro-

jections and different non-planar derivatives as well as

Figure 3: Standard cube map orientation used by our con-

cept and implementation.

A GENERALIZATION APPROACH FOR 3D VIEWING DEFORMATIONS OF SINGLE-CENTER PROJECTIONS

165

= f

sample

)

= f

normal

)

i,j: f

render

)

Tile Screen Feature Map Normal Map Final Rendering

Figure 4: Elaborated conceptual pipeline to render combinations of planar and non-planar projections deﬁned by projection

tiles. A tile screen (A) is transformed into an angle map (B) from which a normal map (C) is derived that contains the cube

map sampling vector to create the ﬁnal rendering (D).

it facilitates the creation of custom projections which

are hard to describe analytically. In this context, pro-

jection tiles are a generalization of the concepts de-

scribed in Section 3. They enable control over differ-

ent projection types as well as add a smooth transition

of their parameters. A projection tile deﬁnes the area

of a speciﬁc projection in relation to the viewport.

Projection tiles are organized in a projection tile

screen (PTS) (Figure 4.A). It is represented by a spe-

ciﬁc feature set T

deﬁned as follows:







··· E







A component E

i j

of T

is denoted as tile feature.

It basically describes a view direction for a speciﬁc

point on the screen. To enable a more intuitive way to

describe a tile screen, we have chosen spherical po-

lar coordinates φ and θ to express the view direction

instead of a normalized direction vector. Thus, we

deﬁne a feature E

i j

as a 6-tupel:

i j

= (x

i j

(φ

i j

),e

(θ

i j

),s

i j

, f

i j

)

i j

, f

i j

∈ [0; 1] ⊂ R

i j

,θ

i j

∈ [−360; 360] ⊂ R

that contains the feature position (x

i j

) and the par-

ticular horizontal and vertical view angles (φ

i j

,θ

i j

The parameter s

i j

is the respective scaling factor while

the variable f

i j

can be used to bind custom parameters

to each feature. Figure 5 demonstrates this by adjust-

ing the image saturation according to the respect value

of f . The function e

(x) = (x + 360)/720 is used to

encode an angle into a range of [0;1].

The PTS is transformed into a feature map (Figure

4.B) by rendering its respective tiles T

with k =

0. ..m −1, l = 0 ... n −1:

= (E

(k,l)

(k+1,l)

(k+1,l+1)

(k,l+1)

)

into a texture using render-to-texture (Wynn, 2002).

The components x

i j

and y

i j

are interpreted as the 2D

vertices of a quad in standard orthographic parallel

projection (Woo et al., 1999). The angles and the

scale factor are encoded into a per-vertex RGBA color

value. The hyperbolic interpolation (Blinn, 1992) be-

tween these values is performed by graphics hard-

ware.

The rendered feature map stores the interpolated hor-

izontal and vertical angles A

= (φ

,θ

) for each

fragment F

. The normal N

can be calculated by:

= f

normal

)

= R



−1

(θ

)









−1

(φ

)























Where R

and R

denote the 3D rotation around the

respective axis. For each N

in the resulting normal

map (Figure 4.C) a sampling vector S can be calcu-

lated by setting δ

·s

) = N

. The result is

shown in Figure 4.D.

5 IMPLEMENTATION

The rendering of non-planar projections is performed

in two steps per frame:

1. Create or update dynamic cube map as described

in Section 5.1. During this phase, optimization

methods as described in Section 5.2 can be uti-

lized.

2. Apply projections by exploiting programmable

graphics hardware (see Section 5.3) This step is

applied in postprocessing and requires an addi-

tional rendering of the scene geometry.

The exemplary implementation was done by us-

ing OpenGL (Segal and Akeley, 2004) in combina-

tion with GLSL (John Kessenich, 2004). It uses

framebuffer objects, ﬂoating point textures, and mip-

mapped cube maps for dynamic texturing (Harris,

2004).

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

166

A B

Figure 5: Examples for combining planar and non-planar projections within a single rendering using projection tile screens.

Sub-ﬁgure A is composed of a planar projection in the center and cylindrical projections left and right. Sub-ﬁgure B shows the

same scene with two planar projections for each cathedral. The saturation falloff is controlled by the respective tile features.

5.1 Cube Map Creation

An advantage of our image-based concept is the appli-

cation of dynamic cube map textures (Greene, 1986).

The creation of those can be optimized depending on

the respective projections. This can be achieved using

two techniques:

Single-Pass Creation: On current hardware, it is

possible to create cube map textures within a single

pass by utilizing geometry shaders (Microsoft, 2007).

This render-to-cube-map technique duplicates each

input triangle six times and applies a separate model-

view transformation for each face of the cube. Each

of the resulting triangles is directed to the appropriate

element of a render target array view of the cube map

(Blythe, 2006)

Multi-Pass Creation: One can create a cube map

texture using multi-pass rendering in combination

with render-to-texture (G

oddeke, 2005). Given a

reference camera position, we can construct six local

virtual cameras with a FOV of 90 degrees, an aspect

ratio of 1, and render the scene into the respective

cube map texture targets.

We have two alternatives to construct these virtual lo-

cal cameras: 1) by rotating the reference camera or 2)

by using ﬁxed cube map orientation (Figure 3). The

latter demands for a look-to correction as described in

Section 3 and is necessary for a simple implementa-

tion of projection tiles.

5.2 Optimization Techniques

Depending on the scene complexity, the creation of a

cube map texture can become costly. Usually, the cre-

ation is ﬁll-limited due to the increased rasterization

effort as well as bandwidth-limited due to the number

of rendering passes. In our application, it is not pos-

sible to use proxy geometry to speed up the creation

process or distribute the rendering of each cube map

face to different frames. We have implemented two

main optimization techniques that are able to com-

pensate this problem.

CMF Optimization: This optimization omits the

update of cube map faces (CMF) that are not visible

in the generated projection. To determine which cube

map faces have to be updated, we calculate 360 de-

gree spherical coordinates for each corner vertex of

the unit cube and the current look-to vector L

. Then,

we deﬁne a non-planar view frustum (NVF) by off-

setting the spherical coordinates of L

with the hori-

zontal and vertical angle of the current projection. If

one of the face vertices is inside the projection region,

the associated face will be updated. Figure 6 demon-

strates this by considering the lower vertices of the

unit cube only. For the current look-to vector L

the

update of CMF

back

is omitted.

(0,0,0)

(1,-1,1)

-z

(-1,-1,1)

(-1,-1,-1) (-1,-1,1)

CMF

back

CMF

front

CMF

left

CMF

right

lower

left

Non-Planar View Frustum (NVF)

Unit Cube

Unit Sphere

lower

right

Projection Region

Figure 6: Parameter space for CMF-Optimization. The

cube map is displayed from the positive y-axis. The gray

area represents the non-planar view frustum of a non-planar

projection.

A GENERALIZATION APPROACH FOR 3D VIEWING DEFORMATIONS OF SINGLE-CENTER PROJECTIONS

167

Figure 7: Example for applying postprocessing ﬁlters to the

output of our projection technique. Together with a wide

angle projection, the radial blur emphasizes the impression

of speed.

Look-To Optimization: Another optimization

method would omit the update of the cube map

texture if the camera position is not changed.

A further optimization method can orient the cube

map in such way that an optimal coverage for the

respective projection region can be achieved. This

includes changes in the projection function and is left

for future work.

5.3 Applying Projections

The projection is rendered in a post-processing pass

subsequent to the cube map creation pass(es). It can

be performed according to the following steps:

1. Setup a standard 2D orthographic parallel projec-

tion with le f t = bottom = 0 and right = top = 1.

The camera is set to the standard orientation with

a look-to vector L

= (0, 0, −1).

2. Activate a speciﬁc fragment program that imple-

ments the mathematical concepts as described in

Sections 3 and 4. Then, the shader program per-

forms cube map texture lookups or outputs the

calculated normal vectors for later re-use.

3. Render a screen aligned quad with standard tex-

ture coordinates that covers the entire screen. To

apply the described concepts, the fragment posi-

tion (s,t) must be normalized in D.

6 RESULTS

Figures 1 and 2 show the application of our approach

for the interactive visualization of large scale datasets

such as 3D virtual city models. Currently, we did not

128,0 256,0 512,0 1024,0

FPS 1 23,3 12,7 8,7 7,8

FPS 2 18,3 10,9 5,1 3,2

FPS 3 9,7 7,7 5,9 4,1

128 256 512 1024

Cube Map Resolution

FPS

Model 1

Model 2

Model 3

Figure 8: Performance measurements for different models

and cube map texture resolutions.

apply our method to scenes made of real-world im-

ages. However, this is possible because our technique

is independent from the scene representation. Fig-

ure 7 demonstrates the combination of our rendering

technique together with color post-processing ﬁlters.

Figure 9 shows the application to projection systems.

Here, the pincushion distortion is compensated by the

projectors.

Performance: Figure 8 shows the measured frame

rate for 3 models of different polygon counts. The

measurements were taken on NVIDIA GeForce 8800

GTX GPU with 786MB RAM and Athlon

62 X2

Dual Core 4200+ with 2.21 GHz and 2 GB of main

memory at a viewport resolution of 1600x1200 pixel.

It uses multi-pass rendering to create the cube map

texture. The test application does not utilize the sec-

ond CPU core. The most complex sample dataset

(Model 1) comprises the inner city of Berlin with

about 16,000 generically textured buildings, about

100 landmarks, and a 3 GB color aerial photo on top

of a digital terrain model.

Limitations: The observable subside of the frame

between the cube map resolution 512 and 1024 pixel

is caused by a ﬁll limitation of the used graphics hard-

ware and can vary. The necessary resolution of the

cube map texture depends on the resolution of the

viewport and the used projection. For high horizon-

tal and vertical FOV, a cube map resolution of 1024

pixels is sufﬁcient in most cases.

The quality of the feature map depends on the res-

olution of the PTS. If the resolution is too low, the

feature interpolation can cause artifacts for tiles with

acute angles. We obtained blurred output images for

extreme FOV. This is the key disadvantage of our ap-

proach and is caused by texture sampling artefacts.

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

168

Figure 9: This is an example for the application of our ren-

dering technique to a cylindrical projection wall with a di-

ameter of 4 meters. It uses a static, pre-rendered cube map

texture.

7 FUTURE WORK

Although our approach performs as expected, there

are still possibilities for further improvements. The

adaption of cube map optimization techniques to the

single pass creation technique as well as experiments

with irregular texture resolutions are left for future

work. To compensate for unsharp output images we

are up to develop custom ﬁltering methods for cube

map textures.

We are particularly interested in supporting other

SCOP projections as described in (Margaret, 1995).

The representation of standard non-planar projections

using tile screens enables the combination of non-

planar projections with the 2D lens metaphor as de-

scribed in (Spindler et al., 2006; Carpendale and

Montagnese, 2001) as well as (Bier et al., 1993).

Therefore, a authoring tool for PTS is necessary. Fur-

ther, we try to adapt our rendering technique to gen-

erate non-planar for 3D anaglyph images.

Additionally, we can apply the isocube approach

(Wan et al., 2007) to compensate cube map sampling

artifacts for lower resolutions. Finally, a comparative

performance evaluation concerning time efﬁciency of

our concept compared to other approaches is left for

future work.

8 CONCLUSIONS

We have presented an approach to generate non-

planar projections for real-time applications. We sug-

gested optimization methods to accelerate the cre-

ation of dynamic environment maps and introduced a

novel approach for creating combined SCOP projec-

tions. We demonstrated the results and performance

of our rendering technique by integrating it into an

existing rendering framework (3Dgeo, 2007).

ACKNOWLEDGEMENTS

This work has been funded by the German Federal

Ministry of Education and Research (BMBF) as part

of the InnoProﬁle research group ’3D Geoinforma-

tion’ (http://www.3dgi.de). The authors would like

to thank also 3D geo for providing the texture data as

well as the cadastral data of virtual 3D city models.

REFERENCES

3Dgeo (2007). LandXplorer. http://www.3dgeo.de.

Agrawala, M., Zorin, D., and Munzner, T. (2000). Artis-

tic Multiprojection Rendering. In 11th Eurographics

Workshop on Rendering, pages 125–136, Brno, Czech

Republic.

Bayarri, S. (1995). Computing Non-Planar Perspectives in

Real Time. Computers & Graphics, 19(3):431–440.

Bier, E. A., Stone, M. C., Pier, K., Buxton, W., and DeRose,

T. D. (1993). Toolglass and Magic Lenses: The

See-Through Interface. In SIGGRAPH, pages 73–80.

ACM Press.

Blinn, J. (1992). Hyperbolic Interpolation. IEEE Computer

Graphics and Applications Staff, 12(4):89–94.

Blythe, D. (2006). The Direct3D 10 System. In SIGGRAPH

’06: ACM SIGGRAPH 2006 Papers, pages 724–734,

New York, NY, USA. ACM Press.

Bourke, P. (2000). Nonlinear Lens Distortion.

Bourke, P. (2004). Offaxis Fisheye Projection.

Brosz, J., Samavati, F. F., Sheelagh, M., Carpendale, T., and

Sousa, M. C. (2007). Single Camera Flexible Pro-

jection. In NPAR ’07: Proceedings of the 5th inter-

national symposium on Non-photorealistic animation

and rendering, pages 33–42, New York, NY, USA.

ACM Press.

Carlbom, I. and Paciorek, J. (1978). Planar Geometric Pro-

jections and Viewing Transformations. ACM Comput.

Surv., 10(4):465–502.

Carpendale, M. S. T. and Montagnese, C. (2001). A Frame-

work for Unifying Presentation Space. In UIST ’01:

Proceedings of the 14th annual ACM symposium on

User interface software and technology, pages 61–70,

New York, NY, USA. ACM Press.

Coleman, P. and Singh, K. (2004). RYAN: Rendering Your

Animation Nonlinearly projected. In NPAR.

Glaeser, G. and Gr

oller, E. (1999). Fast Generation of

Curved Perspectives for Ultra-Wide-Angle Lenses in

VR Applications. The Visual Computer, 15(7/8):365–

376.

Glaeserm, G. (1999). Extreme and Subjective Perspectives.

In Topics in Algebra, Analysis and Geometry, pages

39–51, BPR Mdiatancsad BT/Budapest.

A GENERALIZATION APPROACH FOR 3D VIEWING DEFORMATIONS OF SINGLE-CENTER PROJECTIONS

169

Glasbey, C. and Mardia, K. (1989). A Review of Im-

age Warping Methods. Journal of Applied Statistics,

25:155–171.

Glassner, A. S. (2000). Cubism and Cameras: Free-form

Optics for Computer Graphics. Technical report, Mi-

crosoft Research.

oddeke, D. (2005). Playing Ping Pong with Render-To-

Texture. Technical report, University of Dortmund,

Germany.

Greene, N. (1986). Environment Mapping and other Ap-

plications of World Pojections. IEEE Comput. Graph.

Appl., 6(11):21–29.

oller, M. E. and Acquisto, P. (1993). A Distortion Camera

for Ray Tracing. In Conner, Hernandez, Murthy, and

Power, editors, Visualization and Intelligent Design in

Engineering and Architecture. Elsevier Science Pub-

lishers.

Gustafsson, A. (1993). Interactive Image Warping. Master’s

thesis, Faculty of Information Technology.

H., B., J., Y., and Q., P. (1999). Non-Linear View Interpo-

lation. In The Journal of Visualization and Computer

Animation, volume 10, pages 233–241(9). John Wiley

& Sons, Ltd.

Harris, M. (2004). Dynamic Texturing. NVIDIA Corpora-

tion.

Heidrich, W. and Seidel, H.-P. (1998). View-independent

Environment Maps. In HWWS ’98: Proceedings

of the ACM SIGGRAPH/EUROGRAPHICS workshop

on Graphics hardware, pages 39–ff., New York, NY,

USA. ACM Press.

John Kessenich (2004). The OpenGL Shading Language

Version 1.20, 59 edition.

Kilgard, M. J. (May 19, 2004). NVIDIA OpenGL Exten-

sion Speciﬁcations. Technical report, NVIDIA Cor-

poration.

Leung, Y. and Apperley, M. (1994). A Review and Taxon-

omy of Distortion-Oriented Presentation Techniques.

ACM Transactions on Computer-Human Interaction,

1:126–160.

Margaret, F. (1995). Perspective Projection: the Wrong

Imaging Model.

Microsoft (2007). Direct3D 10 Programming Guide Ex-

cerpts. In SIGGRAPH ’07: ACM SIGGRAPH 2007

courses, pages 369–446, New York, NY, USA. ACM

Press.

Nelson, M. L. (Nicograph 1983). Computer Graphics Dis-

tortion for IMAX and OMNIMAX Projection. In

Nicograph 83MaxNicograph1983, pages 137–159.

Neumann, P. and Carpendale, S. (2003). Taxonomy for Dis-

crete Lenses. Technical Report 2003-734-37, Depart-

ment of Computer Science, University of Calgary.

NVIDIA (2004). OpenGL Cube Map Texturing.

NVIDIA (2005). NVIDIA GPU Programming Guide.

NVIDIA Corporation, 2.4.0 edition.

Polack-Wahl, J. A., Piegl, L. A., and Carter, M. L. (1997).

Perception of Images Using Cylindrical Mapping. The

Visual Computer, 13(4):155–167.

Rademacher, P. and Bishop, G. (1998). Multiple-Center-of-

Projection Images. In SIGGRAPH ’98: Proceedings

of the 25th annual conference on Computer graph-

ics and interactive techniques, pages 199–206, New

York, NY, USA. ACM Press.

Rase, W.-D. (1997). Fischauge-Projektionen als kar-

tographische Lupen. In Dollinger, F. and J.Strobl, ed-

itors, Angewandte Geographische Informationsverar-

beitung, volume IX of Salzburger Geographische Ma-

terialien. Selbstverlag des Instituts f

ur Geographie der

Universit

at Salzburg.

Segal, M. and Akeley, K. (2004). The OpenGL Graphics

System: A Speciﬁcation, Version 2.0.

Singh, K. (2002). A Fresh Perspective. In Graphics Inter-

face, pages 17–24.

Spindler, M., Bubke, M., Germer, T., and Strothotte, T.

(2006). Camera Textures. In GRAPHITE ’06: Pro-

ceedings of the 4th international conference on Com-

puter graphics and interactive techniques in Australa-

sia and Southeast Asia, pages 295–302, New York,

NY, USA. ACM Press.

Swaminathan, R., Grossberg, M., and Nayar, S. (2003).

A Perspective on Distortions. In IEEE Conference

on Computer Vision and Pattern Recognition (CVPR),

volume II, pages 594–601.

Szeliski, R. and Shum, H.-Y. (1997). Creating Full View

Panoramic Image Mosaics and Environment Maps. In

SIGGRAPH ’97: Proceedings of the 24th annual con-

ference on Computer graphics and interactive tech-

niques, pages 251–258, New York, NY, USA. ACM

Press/Addison-Wesley Publishing Co.

Turkowski, K. (1999). Making Environment Maps from

Fisheye Photographs.

van Oortmerssen, W. (2002). FisheyeQuake/PanQuake.

Wan, L., Wong, T.-T., and Leung, C.-S. (2007). Isocube:

Exploiting the Cubemap Hardware. IEEE Trans-

actions on Visualization and Computer Graphics,

13(4):720–731.

Woo, M., Neider, J., Davis, T., and Shreiner, D. (1999).

OpenGL Programming Guide: The Ofﬁcial Guide

to Learning OpenGL, Version 1.2. Addison-Wesley

Longman Publishing Co., Inc., Boston, MA, USA.

Wood, D. N., Finkelstein, A., Hughes, J. F., Thayer, C. E.,

and Salesin, D. H. (1997). Multiperspective Panora-

mas for Cel Animation. In SIGGRAPH ’97: Pro-

ceedings of the 24th annual conference on Computer

graphics and interactive techniques, pages 243–250,

New York, NY, USA. ACM Press/Addison-Wesley

Publishing Co.

Wynn, C. (2002). OpenGL Render-to-Texture. In GDC.

NVIDIA Corporation.

Yang, Y., Chen, J. X., and Beheshti, M. (2005). Nonlinear

Perspective Projections and Magic Lenses: 3D View

Deformation. IEEE Computer Graphics and Applica-

tions, pages 76–84.

Yang, Y., Chen, J. X., Kim, W., , and Kee, C. (2003). Non-

linear Pojection: Using Deformations in 3D Viewing.

In Jim X. Chen, editor, Visualization Corner, pages

54–59. IEEE.

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

170