SlideShare a Scribd company logo
Improved Alpha-Tested Magnification for Vector Textures and Special Effects 
Chris Green 
Valve 
(a) 64x64 texture, alpha-blended (b) 64x64 texture, alpha tested (c) 64x64 texture using our technique 
Figure 1: Vector art encoded in a 64x64 texture using (a) simple bilinear filtering (b) alpha testing and (c) our distance field technique 
Abstract 
A simple and efficient method is presented which allows improved 
rendering of glyphs composed of curved and linear elements. A 
distance field is generated from a high resolution image, and then 
stored into a channel of a lower-resolution texture. In the simplest 
case, this texture can then be rendered simply by using the alpha-testing 
and alpha-thresholding feature of modern GPUs, without a 
custom shader. This allows the technique to be used on even the 
lowest-end 3D graphics hardware. 
With the use of programmable shading, the technique is extended 
to perform various special effect renderings, including soft edges, 
outlining, drop shadows, multi-colored images, and sharp corners. 
1 Introduction 
For high quality real-time 3D rendering, it is critical that the limited 
amount of memory available for the storage of texture maps be used 
efficiently. In interactive applications such as computer games, the 
user is often able to view texture mapped objects at a high level 
of magnification, requiring that texture maps be stored at a high 
resolution so as to not become unpleasantly blurry, as shown in 
Figure 1a, when viewed from such perspectives. 
When the texture maps are used to represent “line-art” images, such 
as text, signs and UI elements, this can require the use of very high 
resolution texture maps in order to look acceptable, particularly at 
high resolutions. 
In addition to text and UI elements, this problem is also common 
in alpha-tested image-based impostors for complicated objects such 
as foliage. When textures with alpha channels derived from cover-age 
are magnified with hardware bilinear filtering, unpleasant “wig-gles” 
as seen in Figure 1b appear because the coverage function is 
not linear. 
In this chapter, we present a simple method to generate and ren-der 
alpha-tested texture maps in order to minimize the artifacts that 
arise when such textures are heavily magnified. We will demon-strate 
several usage scenarios in a computer game context, both for 
e-mail: cgreen@valvesoftware.com 
3D renderings and also user-interface elements. Our technique is 
capable of generating high quality vector art renderings as shown 
in Figure 1c. 
2 Related work 
Many techniques have been developed to accurately render vec-tor 
graphics using texture-mapping graphics hardware. In [Frisken 
et al. 2000], distance fields were used to represent both 2- 
dimensional glyphs and 3-dimensional solid geometry. Quadtrees 
and octrees were used to adaptively control the resolution of the 
distance field based upon local variations. While GPU rendering 
of such objects was not discussed, recent advances in the general-ity 
of GPU programming models would allow this method to be 
implemented using DirectX10 [Blythe 2006]. 
In [Sen 2004] and [Tumblin and Choudhury 2004], texture maps 
are augmented with additional data to control interpolation between 
texel samples so as to add sharp edges in a controllable fashion. 
Both line-art images and photographic textures containing hard 
edges were rendered directly on the GPU using their representa-tion. 
In [Loop and Blinn 2005], implicit cubic curves were used to model 
the boundaries of glyphs, with the GPU used to render vector tex-tures 
with smooth resolution-independent curves. 
In [Qin et al. 2006], a distance based representation is used, with 
a precomputed set of “features” influencing each Voronoi region. 
Given these features, a pixel shader is used to analytically compute 
exact distance values. 
Alpha-testing, in which the alpha value output from the pixel shader 
is thresholded so as to yield a binary on/off result, is widely used 
in games to provide sharp edges in reconstructed textures. Unfortu-nately, 
because the images that are generally used as sources for this 
contain “coverage” information which is not properly reconstructed 
at the subtexel level by bilinear interpolation, unpleasant artifacts 
are usually seen for non-axis-aligned edges when these textures are 
magnified.
4 RENDERING 
3 Representation and Generation 
In order to overcome the artifacts of simple alpha testing while 
keeping storage increase to a minimum, we sought a method for 
displaying vector textures that could 
• work on all levels of graphics hardware, including systems 
lacking programmable shading 
• run as fast as, or nearly as fast as, standard texture mapping 
• take advantage of the bilinear interpolation present in all mod-ern 
GPUs 
• function inside of a pre-existing complex shader sys-tem 
[Mitchell et al. 2006] with few changes 
• add at most a few instructions to the pixel shader so that vector 
textures can be used in existing shaders without overflowing 
instruction limits 
• not require that input images be provided in a vector form. 
• use existing low-precision 8-bit texture formats 
• be used as a direct replacement for alpha-tested impostor im-ages 
We chose to implement a simple uniformly-sampled signed-distance 
field representation, with the distance function stored in 
an 8-bit channel. By doing so, we are able to take advantage of 
the native bilinear texture interpolation which is present in all mod-ern 
GPUs in order to accurately reconstruct the distance between 
a sub-texel and a piecewise-linear approximation of the true high-resolution 
image. While this representation is limited in terms of 
the topology of features which can be represented compared to 
other approaches, we felt that its high performance, simplicity, and 
ease of integration with our existing rendering system made it the 
right choice for Valve’s Source engine. 
While it is possible to generate the appropriate distance data from 
vector-based representations of the underlying art, we choose in-stead 
to generate the low-resolution distance fields from high reso-lution 
source images. In a typical case, a 4096x4096 image will be 
used to generate a distance field texture with a resolution as low as 
64x64, as shown in Figure 2. 
At texture-generation time, the generator takes as its input a high 
resolution binary texture where each texel is classified as either “in” 
or “out.” The user specifies a target resolution, and also a “spread 
factor,” which controls the range which is used to map the signed 
distance into the range of 0 to 1 for storage in an 8-bit texture chan-nel. 
The spread factor also controls the domain of effect for such 
special rendering attributes as drop-shadows and outlines, which 
will be discussed in section 4.2. 
For each output texel, the distance field generator determines 
whether the corresponding pixel in the high resolution image is “in” 
or “out.” Additionally, the generator computes 2D distance (in tex-els) 
to the nearest texel of the opposite state. This is done by ex-amining 
the local neighborhood around a given texel. While there 
are more efficient and complex algorithms to compute the signed 
distance field than our simple “brute-force” search, because of the 
limited distance range which may be stored in an 8-bit alpha chan-nel, 
only a small neighborhood must be searched. The execution 
time for this simple brute-force method is negligible. 
Once this signed distance has been calculated, we map it into the 
range 0..1, with 0 representing the maximum possible negative dis-tance 
and 1.0 representing the maximum possible positive distance. 
A texel value of 0.5 represents the exact position of the edge and, 
hence, 0.5 is generally used for the alpha threshold value. 
(a) High resolution input (b) 64x64 Distance field 
Figure 2: (a) A high resolution (4096x4096) binary input is used 
to compute (b) a low resolution (64x64) distance field 
4 Rendering 
In the simplest case, the resulting distance field textures can be used 
as-is in any context where geometry is being rendered with alpha-testing. 
Under magnification, this will produce an image with high-resolution 
(albeit, aliased) linear edges, free of the false curved con-tours 
(see Figure 1b) common with alpha-tested textures generated 
by storing and filtering coverage rather than a distance field. With 
a distance field representation, we merely have to set the alpha test 
threshold to 0.5. Since it is fairly common to use alpha testing 
rather than alpha blending for certain classes of primitives in or-der 
to avoid the costly sorting step, this technique can provide an 
immediate visual improvement with no performance penalty. 
Figure 3: 128x128 “No trespassing” distance image applied to a 
surface in Team Fortress 2 
In Figure 3, we demonstrate a 128x128 distance field representation 
of a “No Trespassing” sign rendered as a decal over the surface of 
a wall in the game Team Fortress 2. The apparent resolution of this 
decal is incredibly high in world space and holds up well under any 
level of magnification that it will ever undergo in the game. We 
will refer to this particular decal example in the next section as we 
discuss other enhancements available to us when representing our 
vector art in this manner. 
4.1 Antialiasing 
If alpha-blending is practical for a given application, the same dis-tance 
field representation can be used to generate higher quality 

c 2007 Valve Corporation. All Rights Reserved
4.3 Sharp corners 4 RENDERING 
Figure 4: Zoom of 256x256 “No Trespassing” sign with hard (left) and softened edges (right) 
renderings than mere alpha testing, at the expense of requiring cus-tom 
fragment shaders. 
Figure 4 demonstrates a simple way to soften up the harsh aliased 
pixel edges. Two distance thresholds, Distmin and Distmax, are 
defined and the shader maps the distance field value between these 
two values using the smoothstep() function. On graphics hardware 
which supports per-pixel screen-space derivatives, the derivatives of 
the distance field’s texture coordinates can be used to vary the width 
of the soft region in order to properly anti-alias the edges of the vec-tor 
art [Qin et al. 2006]. When the texture is minified, widening of 
the soft region can be used to reduce aliasing artifacts. Addition-ally, 
when rendering alpha-tested foliage, the alpha threshold can 
be increased with distance, so that the foliage gradually disappears 
as it becomes farther away to avoid LOD popping. 
4.2 Enhanced Rendering 
In addition to providing crisp high resolution antialiased vector art 
using raster hardware, we can apply additional manipulations using 
the distance field to achieve other effects such as outlining, glows 
and drop shadows. Of course, since all of these operations are func-tions 
of the distance field, they can be dynamically controlled with 
shader parameters. 
4.2.1 Outlining 
By changing the color of all texels which are between two user-specified 
distance values, a simple texture-space outlining can be 
applied by the pixel shader as shown in our decal example in Fig-ure 
5. The outline produced will have crisp high quality edges when 
magnified and, of course, the color and width of the outline can be 
varied dynamically merely by changing pixel shader constants. 
4.2.2 Glows 
When the alpha value is between the threshold value of 0.5 and 0, 
the smoothstep function can be used to substitute a “halo” whose 
color value comes from a pixel shader constant as shown in Fig-ure 
6. The dynamic nature of this effect is particularly powerful 
in a game, as designers may want to draw attention to a particular 
piece of vector art in the game world based on some game state by 
animating the glow parameters (blinking a health meter, emphasiz-ing 
an exit sign etc). 
Figure 5: Outline added by pixel shader 
4.2.3 Drop Shadows 
In addition to effects which are simple functions of a single dis-tance, 
we can use a second lookup into the distance field with a 
texture coordinate offset to produce drop shadows or other similar 
effects as shown in Figure 7. 
In addition to these simple 2D effects, there are surely other ways 
to reinterpret the distance field to give designers even more options. 
4.3 Sharp corners 
As you may have noticed in all of the preceding examples, encoding 
edges using a single signed distance “rounds off” corners as the 
resolution of the distance field decreases [Qin et al. 2006]. For 
example, the hard corners of the letter G in Figure 2a become more 
rounded off as illustrated in Figures 5, 6 and 7. 
Sharp corners can be preserved, however, by using more than 
one channel of the texture to represent different edges intersect-ing 
within a texel. For instance, with two channels, the intersection 
of two edges can be accurately represented by performing a logical 
AND in the pixel shader. In Figure 8, we have stored these two edge 
distances in the red and green channels of a single texture, resulting 
in a well-preserved pointy corner. This same technique could also 
be performed on the “No Trespassing” sign if we wished to repre-sent 
sharper corners on our text. As it stands, we like the rounded 
style of this text and have used a single distance field for this and 
other decals in Team Fortress 2. 

c 2007 Valve Corporation. All Rights Reserved
REFERENCES REFERENCES 
Figure 6: Scary flashing “Outer glow” added by pixel shader 
Figure 7: Soft drop-shadow added by the pixel shader. The direc-tion, 
size, opacity, and color of the shadow are dynamically con-trollable. 
5 Conclusion 
In this chapter, we have demonstrated an efficient vector texture 
system which has been integrated into the Source game engine 
which has been previously used to develop games such as the Half- 
Life 2 series, Counter-Strike: Source and Day of Defeat: Source. 
This vector texture technology is used in the upcoming game Team 
Fortress 2 with no significant performance degradation relative to 
conventional texture mapping. We were able to effectively use 
vector-encoded images both for textures mapped onto 3D geom-etry 
in our first person 3D view and also for 2D screen overlays. 
This capability has provided significant visual improvements and 
savings of texture memory. 
References 
BLYTHE, D. 2006. The direct3d 10 system. In SIGGRAPH ’06: 
ACM SIGGRAPH 2006 Papers, ACM Press, New York, NY, 
USA, 724–734. 
FRISKEN, S. F., PERRY, R. N., ROCKWOOD, A. P., AND JONES, 
T. R. 2000. Adaptively sampled distance fields: a general 
representation of shape for computer graphics. In SIGGRAPH 
’00: Proceedings of the 27th annual conference on Com-puter 
graphics and interactive techniques, ACM Press/Addison- 
Wesley Publishing Co., New York, NY, USA, 249–254. 
LOOP, C., AND BLINN, J. 2005. Resolution independent curve 
rendering using programmable graphics hardware. In SIG-f 
l o a t di s tAlphaMa sk = b a s eCo l o r . a ; i f ( OUTLINE  
( di s tAlphaMa sk = OUTLINE MIN VALUE0 )  
( di s tAlphaMa sk = OUTLINE MAX VALUE1 ) ) 
{ 
f l o a t o F a c t o r = 1 . 0 ; 
i f ( di s tAlphaMa sk = OUTLINE MIN VALUE1 ) 
{ 
o F a c t o r = smo o t h s t e p ( OUTLINE MIN VALUE0 , 
OUTLINE MIN VALUE1 , 
di s tAlphaMa sk ) ; 
} 
e l s e 
{ 
o F a c t o r = smo o t h s t e p ( OUTLINE MAX VALUE1 , 
OUTLINE MAX VALUE0 , 
di s tAlphaMa sk ) ; 
} 
b a s eCo l o r = l e r p ( ba s eColor , OUTLINE COLOR, o F a c t o r ) ; 
} 
i f ( SOFT EDGES ) { 
b a s eCo l o r . a = smo o t h s t e p ( SOFT EDGE MIN, 
SOFT EDGE MAX, 
di s tAlphaMa sk ) ; 
} e l s e { 
b a s eCo l o r . a = di s tAlphaMa sk = 0 . 5 ; 
} i f ( OUTER GLOW ) { 
f l o a t 4 glowTexe l = 
tex2D ( Ba s eTextur eSampl e r , 
i . baseTexCoord . xy+GLOW UV OFFSET ) ; 
f l o a t 4 glowc = OUTER GLOW COLOR  smo o t h s t e p ( 
OUTER GLOW MIN DVALUE, 
OUTER GLOW MAX DVALUE, 
glowTexe l . a ) ; 
b a s eCo l o r = l e r p ( glowc , ba s eColor , mskUsed ) ; 
} 
Figure 9: HLSL source code for outline, glow/dropshadow, and 
edge softness. 
GRAPH ’05: ACM SIGGRAPH 2005 Papers, ACM Press, New 
York, NY, USA, 1000–1009. 
MITCHELL, J., MCTAGGART, G., AND GREEN, C. 2006. Shading 
in valve’s source engine. In SIGGRAPH ’06: ACM SIGGRAPH 
2006 Courses, ACM Press, New York, NY, USA, 129–142. 
QIN, Z., MCCOOL, M. D., AND KAPLAN, C. S. 2006. Real-time 
texture-mapped vector glyphs. In I3D ’06: Proceedings of the 
2006 symposium on Interactive 3D graphics and games, ACM 
Press, New York, NY, USA, 125–132. 
RAY, N., NEIGER, T., CAVIN, X., AND LEVY, B. 2005. Vector 
texture maps. In Tech Report. 
SEN, P. 2004. Silhouette maps for improved texture magnifica-tion. 
In HWWS ’04: Proceedings of the ACM SIGGRAPH/EU-ROGRAPHICS 
conference on Graphics hardware, ACM Press, 
New York, NY, USA, 65–73. 
TUMBLIN, J., AND CHOUDHURY, P. 2004. Bixels: Picture sam-ples 
with sharp embedded boundaries. In Rendering Techniques, 
255–264. 

c 2007 Valve Corporation. All Rights Reserved
REFERENCES REFERENCES 
Figure 8: Corner encoded at 64x64 using one distance field (left) and the AND of two distance fields (right) 

c 2007 Valve Corporation. All Rights Reserved

More Related Content

PPTX
Image representation
PDF
Image Segmentation
PPTX
Comparison of image segmentation
PPTX
LAPLACE TRANSFORM SUITABILITY FOR IMAGE PROCESSING
PPTX
Image parts and segmentation
PPT
Image inpainting
PPT
Image processing
PPT
Image Texture Analysis
Image representation
Image Segmentation
Comparison of image segmentation
LAPLACE TRANSFORM SUITABILITY FOR IMAGE PROCESSING
Image parts and segmentation
Image inpainting
Image processing
Image Texture Analysis

What's hot (20)

PPT
Chapter10 image segmentation
PPTX
Image segmentation
PPTX
Image Representation & Descriptors
PPT
Image enhancement
PDF
4 image enhancement in spatial domain
PPTX
Segmentation Techniques -II
PPTX
Image enhancement techniques
PPTX
Image enhancement
PDF
Digital image processing using matlab: basic transformations, filters and ope...
PDF
Contrast enhancement in digital images
PPT
Image segmentation
PPT
Image enhancement in the spatial domain1
PPTX
Image segmentation
DOCX
WBOIT Final Version
PPT
regions
PPTX
Histogram Processing
PPT
Image segmentation ppt
PPS
Icdecs 2011
PPTX
Segmentation Techniques -I
Chapter10 image segmentation
Image segmentation
Image Representation & Descriptors
Image enhancement
4 image enhancement in spatial domain
Segmentation Techniques -II
Image enhancement techniques
Image enhancement
Digital image processing using matlab: basic transformations, filters and ope...
Contrast enhancement in digital images
Image segmentation
Image enhancement in the spatial domain1
Image segmentation
WBOIT Final Version
regions
Histogram Processing
Image segmentation ppt
Icdecs 2011
Segmentation Techniques -I
Ad

Viewers also liked (11)

PDF
Possibilities of non commercial game
PDF
Japanese web browser preferences | 世論 what japan thinks
PDF
Success stories compilation of game-based learning initiatives in adults educ...
PDF
Bullet Physic Engine SDK
PDF
Playability and Player Experience Research
PDF
CEO Survey Report of Japanese Video Game Developers - Mirko Ernkvist
PDF
Smoothed Particle Hydrodynamics
PDF
Application parallelisation Android - Klaas Vangend
PDF
AI in Games- Steering, Wander and Flocking behavior
PDF
Getting Native with NDK
PDF
Children's Books, Stories and Songs - Children literature
Possibilities of non commercial game
Japanese web browser preferences | 世論 what japan thinks
Success stories compilation of game-based learning initiatives in adults educ...
Bullet Physic Engine SDK
Playability and Player Experience Research
CEO Survey Report of Japanese Video Game Developers - Mirko Ernkvist
Smoothed Particle Hydrodynamics
Application parallelisation Android - Klaas Vangend
AI in Games- Steering, Wander and Flocking behavior
Getting Native with NDK
Children's Books, Stories and Songs - Children literature
Ad

Similar to Improved Alpha-Tested Magnification for Vector Textures and Special Effects (20)

PDF
A0280105
PDF
Automatic Detection of Window Regions in Indoor Point Clouds Using R-CNN
PDF
stylegan.pdf
PDF
Road Segmentation from satellites images
PDF
IRJET- An Approach to FPGA based Implementation of Image Mosaicing using Neur...
PDF
A PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKING
PDF
Wordoku Puzzle Solver - Image Processing Project
PDF
JonathanWestlake_ComputerVision_Project1
PPTX
Single Image Depth Estimation using frequency domain analysis and Deep learning
PPT
Advanced Lighting for Interactive Applications
PPT
Signature recognition using clustering techniques dissertati
PDF
Fractal Image Compression of Satellite Color Imageries Using Variable Size of...
PDF
An Analysis of Energy Efficient Gaussian Filter Architectures
PDF
Developing 3D Viewing Model from 2D Stereo Pair with its Occlusion Ratio
DOCX
Assignment-1-NF.docx
PPTX
2022-01-17-Rethinking_Bisenet.pptx
PPTX
Unit 2 open gl .pptx
PDF
License plate recognition
PPTX
Normal Mapping / Computer Graphics - IK
PDF
Survey paper on image compression techniques
A0280105
Automatic Detection of Window Regions in Indoor Point Clouds Using R-CNN
stylegan.pdf
Road Segmentation from satellites images
IRJET- An Approach to FPGA based Implementation of Image Mosaicing using Neur...
A PROJECT REPORT ON REMOVAL OF UNNECESSARY OBJECTS FROM PHOTOS USING MASKING
Wordoku Puzzle Solver - Image Processing Project
JonathanWestlake_ComputerVision_Project1
Single Image Depth Estimation using frequency domain analysis and Deep learning
Advanced Lighting for Interactive Applications
Signature recognition using clustering techniques dissertati
Fractal Image Compression of Satellite Color Imageries Using Variable Size of...
An Analysis of Energy Efficient Gaussian Filter Architectures
Developing 3D Viewing Model from 2D Stereo Pair with its Occlusion Ratio
Assignment-1-NF.docx
2022-01-17-Rethinking_Bisenet.pptx
Unit 2 open gl .pptx
License plate recognition
Normal Mapping / Computer Graphics - IK
Survey paper on image compression techniques

More from ナム-Nam Nguyễn (10)

PDF
Unite 2013 optimizing unity games for mobile platforms
PDF
China mobile games and entertainment group
PDF
Abusing mobilegames
PDF
MGI Big data full report
PDF
Ai.game.wisdom
PDF
Physics for Game Programmers: Spatial Data Structures
PDF
The Economics of the “China Price”
PDF
Android Lifecycle Basics in Practice.
PDF
Practical Source Code for Ship Motions Time Domain Numerical Analysis and Its...
PDF
Luoc su duc phat thich nu hue ngan core- da chuyen
Unite 2013 optimizing unity games for mobile platforms
China mobile games and entertainment group
Abusing mobilegames
MGI Big data full report
Ai.game.wisdom
Physics for Game Programmers: Spatial Data Structures
The Economics of the “China Price”
Android Lifecycle Basics in Practice.
Practical Source Code for Ship Motions Time Domain Numerical Analysis and Its...
Luoc su duc phat thich nu hue ngan core- da chuyen

Recently uploaded (20)

PPTX
FINAL REVIEW FOR COPD DIANOSIS FOR PULMONARY DISEASE.pptx
PDF
Operating System & Kernel Study Guide-1 - converted.pdf
PPTX
OOP with Java - Java Introduction (Basics)
PDF
Mitigating Risks through Effective Management for Enhancing Organizational Pe...
PPTX
Lesson 3_Tessellation.pptx finite Mathematics
DOCX
573137875-Attendance-Management-System-original
PPT
Project quality management in manufacturing
PPTX
UNIT 4 Total Quality Management .pptx
DOCX
ASol_English-Language-Literature-Set-1-27-02-2023-converted.docx
PPTX
Internet of Things (IOT) - A guide to understanding
PPT
Mechanical Engineering MATERIALS Selection
PDF
Well-logging-methods_new................
PPTX
CH1 Production IntroductoryConcepts.pptx
PDF
PPT on Performance Review to get promotions
PPTX
web development for engineering and engineering
PPTX
additive manufacturing of ss316l using mig welding
PDF
Digital Logic Computer Design lecture notes
PPTX
bas. eng. economics group 4 presentation 1.pptx
PPTX
M Tech Sem 1 Civil Engineering Environmental Sciences.pptx
PDF
Structs to JSON How Go Powers REST APIs.pdf
FINAL REVIEW FOR COPD DIANOSIS FOR PULMONARY DISEASE.pptx
Operating System & Kernel Study Guide-1 - converted.pdf
OOP with Java - Java Introduction (Basics)
Mitigating Risks through Effective Management for Enhancing Organizational Pe...
Lesson 3_Tessellation.pptx finite Mathematics
573137875-Attendance-Management-System-original
Project quality management in manufacturing
UNIT 4 Total Quality Management .pptx
ASol_English-Language-Literature-Set-1-27-02-2023-converted.docx
Internet of Things (IOT) - A guide to understanding
Mechanical Engineering MATERIALS Selection
Well-logging-methods_new................
CH1 Production IntroductoryConcepts.pptx
PPT on Performance Review to get promotions
web development for engineering and engineering
additive manufacturing of ss316l using mig welding
Digital Logic Computer Design lecture notes
bas. eng. economics group 4 presentation 1.pptx
M Tech Sem 1 Civil Engineering Environmental Sciences.pptx
Structs to JSON How Go Powers REST APIs.pdf

Improved Alpha-Tested Magnification for Vector Textures and Special Effects

  • 1. Improved Alpha-Tested Magnification for Vector Textures and Special Effects Chris Green Valve (a) 64x64 texture, alpha-blended (b) 64x64 texture, alpha tested (c) 64x64 texture using our technique Figure 1: Vector art encoded in a 64x64 texture using (a) simple bilinear filtering (b) alpha testing and (c) our distance field technique Abstract A simple and efficient method is presented which allows improved rendering of glyphs composed of curved and linear elements. A distance field is generated from a high resolution image, and then stored into a channel of a lower-resolution texture. In the simplest case, this texture can then be rendered simply by using the alpha-testing and alpha-thresholding feature of modern GPUs, without a custom shader. This allows the technique to be used on even the lowest-end 3D graphics hardware. With the use of programmable shading, the technique is extended to perform various special effect renderings, including soft edges, outlining, drop shadows, multi-colored images, and sharp corners. 1 Introduction For high quality real-time 3D rendering, it is critical that the limited amount of memory available for the storage of texture maps be used efficiently. In interactive applications such as computer games, the user is often able to view texture mapped objects at a high level of magnification, requiring that texture maps be stored at a high resolution so as to not become unpleasantly blurry, as shown in Figure 1a, when viewed from such perspectives. When the texture maps are used to represent “line-art” images, such as text, signs and UI elements, this can require the use of very high resolution texture maps in order to look acceptable, particularly at high resolutions. In addition to text and UI elements, this problem is also common in alpha-tested image-based impostors for complicated objects such as foliage. When textures with alpha channels derived from cover-age are magnified with hardware bilinear filtering, unpleasant “wig-gles” as seen in Figure 1b appear because the coverage function is not linear. In this chapter, we present a simple method to generate and ren-der alpha-tested texture maps in order to minimize the artifacts that arise when such textures are heavily magnified. We will demon-strate several usage scenarios in a computer game context, both for e-mail: cgreen@valvesoftware.com 3D renderings and also user-interface elements. Our technique is capable of generating high quality vector art renderings as shown in Figure 1c. 2 Related work Many techniques have been developed to accurately render vec-tor graphics using texture-mapping graphics hardware. In [Frisken et al. 2000], distance fields were used to represent both 2- dimensional glyphs and 3-dimensional solid geometry. Quadtrees and octrees were used to adaptively control the resolution of the distance field based upon local variations. While GPU rendering of such objects was not discussed, recent advances in the general-ity of GPU programming models would allow this method to be implemented using DirectX10 [Blythe 2006]. In [Sen 2004] and [Tumblin and Choudhury 2004], texture maps are augmented with additional data to control interpolation between texel samples so as to add sharp edges in a controllable fashion. Both line-art images and photographic textures containing hard edges were rendered directly on the GPU using their representa-tion. In [Loop and Blinn 2005], implicit cubic curves were used to model the boundaries of glyphs, with the GPU used to render vector tex-tures with smooth resolution-independent curves. In [Qin et al. 2006], a distance based representation is used, with a precomputed set of “features” influencing each Voronoi region. Given these features, a pixel shader is used to analytically compute exact distance values. Alpha-testing, in which the alpha value output from the pixel shader is thresholded so as to yield a binary on/off result, is widely used in games to provide sharp edges in reconstructed textures. Unfortu-nately, because the images that are generally used as sources for this contain “coverage” information which is not properly reconstructed at the subtexel level by bilinear interpolation, unpleasant artifacts are usually seen for non-axis-aligned edges when these textures are magnified.
  • 2. 4 RENDERING 3 Representation and Generation In order to overcome the artifacts of simple alpha testing while keeping storage increase to a minimum, we sought a method for displaying vector textures that could • work on all levels of graphics hardware, including systems lacking programmable shading • run as fast as, or nearly as fast as, standard texture mapping • take advantage of the bilinear interpolation present in all mod-ern GPUs • function inside of a pre-existing complex shader sys-tem [Mitchell et al. 2006] with few changes • add at most a few instructions to the pixel shader so that vector textures can be used in existing shaders without overflowing instruction limits • not require that input images be provided in a vector form. • use existing low-precision 8-bit texture formats • be used as a direct replacement for alpha-tested impostor im-ages We chose to implement a simple uniformly-sampled signed-distance field representation, with the distance function stored in an 8-bit channel. By doing so, we are able to take advantage of the native bilinear texture interpolation which is present in all mod-ern GPUs in order to accurately reconstruct the distance between a sub-texel and a piecewise-linear approximation of the true high-resolution image. While this representation is limited in terms of the topology of features which can be represented compared to other approaches, we felt that its high performance, simplicity, and ease of integration with our existing rendering system made it the right choice for Valve’s Source engine. While it is possible to generate the appropriate distance data from vector-based representations of the underlying art, we choose in-stead to generate the low-resolution distance fields from high reso-lution source images. In a typical case, a 4096x4096 image will be used to generate a distance field texture with a resolution as low as 64x64, as shown in Figure 2. At texture-generation time, the generator takes as its input a high resolution binary texture where each texel is classified as either “in” or “out.” The user specifies a target resolution, and also a “spread factor,” which controls the range which is used to map the signed distance into the range of 0 to 1 for storage in an 8-bit texture chan-nel. The spread factor also controls the domain of effect for such special rendering attributes as drop-shadows and outlines, which will be discussed in section 4.2. For each output texel, the distance field generator determines whether the corresponding pixel in the high resolution image is “in” or “out.” Additionally, the generator computes 2D distance (in tex-els) to the nearest texel of the opposite state. This is done by ex-amining the local neighborhood around a given texel. While there are more efficient and complex algorithms to compute the signed distance field than our simple “brute-force” search, because of the limited distance range which may be stored in an 8-bit alpha chan-nel, only a small neighborhood must be searched. The execution time for this simple brute-force method is negligible. Once this signed distance has been calculated, we map it into the range 0..1, with 0 representing the maximum possible negative dis-tance and 1.0 representing the maximum possible positive distance. A texel value of 0.5 represents the exact position of the edge and, hence, 0.5 is generally used for the alpha threshold value. (a) High resolution input (b) 64x64 Distance field Figure 2: (a) A high resolution (4096x4096) binary input is used to compute (b) a low resolution (64x64) distance field 4 Rendering In the simplest case, the resulting distance field textures can be used as-is in any context where geometry is being rendered with alpha-testing. Under magnification, this will produce an image with high-resolution (albeit, aliased) linear edges, free of the false curved con-tours (see Figure 1b) common with alpha-tested textures generated by storing and filtering coverage rather than a distance field. With a distance field representation, we merely have to set the alpha test threshold to 0.5. Since it is fairly common to use alpha testing rather than alpha blending for certain classes of primitives in or-der to avoid the costly sorting step, this technique can provide an immediate visual improvement with no performance penalty. Figure 3: 128x128 “No trespassing” distance image applied to a surface in Team Fortress 2 In Figure 3, we demonstrate a 128x128 distance field representation of a “No Trespassing” sign rendered as a decal over the surface of a wall in the game Team Fortress 2. The apparent resolution of this decal is incredibly high in world space and holds up well under any level of magnification that it will ever undergo in the game. We will refer to this particular decal example in the next section as we discuss other enhancements available to us when representing our vector art in this manner. 4.1 Antialiasing If alpha-blending is practical for a given application, the same dis-tance field representation can be used to generate higher quality c 2007 Valve Corporation. All Rights Reserved
  • 3. 4.3 Sharp corners 4 RENDERING Figure 4: Zoom of 256x256 “No Trespassing” sign with hard (left) and softened edges (right) renderings than mere alpha testing, at the expense of requiring cus-tom fragment shaders. Figure 4 demonstrates a simple way to soften up the harsh aliased pixel edges. Two distance thresholds, Distmin and Distmax, are defined and the shader maps the distance field value between these two values using the smoothstep() function. On graphics hardware which supports per-pixel screen-space derivatives, the derivatives of the distance field’s texture coordinates can be used to vary the width of the soft region in order to properly anti-alias the edges of the vec-tor art [Qin et al. 2006]. When the texture is minified, widening of the soft region can be used to reduce aliasing artifacts. Addition-ally, when rendering alpha-tested foliage, the alpha threshold can be increased with distance, so that the foliage gradually disappears as it becomes farther away to avoid LOD popping. 4.2 Enhanced Rendering In addition to providing crisp high resolution antialiased vector art using raster hardware, we can apply additional manipulations using the distance field to achieve other effects such as outlining, glows and drop shadows. Of course, since all of these operations are func-tions of the distance field, they can be dynamically controlled with shader parameters. 4.2.1 Outlining By changing the color of all texels which are between two user-specified distance values, a simple texture-space outlining can be applied by the pixel shader as shown in our decal example in Fig-ure 5. The outline produced will have crisp high quality edges when magnified and, of course, the color and width of the outline can be varied dynamically merely by changing pixel shader constants. 4.2.2 Glows When the alpha value is between the threshold value of 0.5 and 0, the smoothstep function can be used to substitute a “halo” whose color value comes from a pixel shader constant as shown in Fig-ure 6. The dynamic nature of this effect is particularly powerful in a game, as designers may want to draw attention to a particular piece of vector art in the game world based on some game state by animating the glow parameters (blinking a health meter, emphasiz-ing an exit sign etc). Figure 5: Outline added by pixel shader 4.2.3 Drop Shadows In addition to effects which are simple functions of a single dis-tance, we can use a second lookup into the distance field with a texture coordinate offset to produce drop shadows or other similar effects as shown in Figure 7. In addition to these simple 2D effects, there are surely other ways to reinterpret the distance field to give designers even more options. 4.3 Sharp corners As you may have noticed in all of the preceding examples, encoding edges using a single signed distance “rounds off” corners as the resolution of the distance field decreases [Qin et al. 2006]. For example, the hard corners of the letter G in Figure 2a become more rounded off as illustrated in Figures 5, 6 and 7. Sharp corners can be preserved, however, by using more than one channel of the texture to represent different edges intersect-ing within a texel. For instance, with two channels, the intersection of two edges can be accurately represented by performing a logical AND in the pixel shader. In Figure 8, we have stored these two edge distances in the red and green channels of a single texture, resulting in a well-preserved pointy corner. This same technique could also be performed on the “No Trespassing” sign if we wished to repre-sent sharper corners on our text. As it stands, we like the rounded style of this text and have used a single distance field for this and other decals in Team Fortress 2. c 2007 Valve Corporation. All Rights Reserved
  • 4. REFERENCES REFERENCES Figure 6: Scary flashing “Outer glow” added by pixel shader Figure 7: Soft drop-shadow added by the pixel shader. The direc-tion, size, opacity, and color of the shadow are dynamically con-trollable. 5 Conclusion In this chapter, we have demonstrated an efficient vector texture system which has been integrated into the Source game engine which has been previously used to develop games such as the Half- Life 2 series, Counter-Strike: Source and Day of Defeat: Source. This vector texture technology is used in the upcoming game Team Fortress 2 with no significant performance degradation relative to conventional texture mapping. We were able to effectively use vector-encoded images both for textures mapped onto 3D geom-etry in our first person 3D view and also for 2D screen overlays. This capability has provided significant visual improvements and savings of texture memory. References BLYTHE, D. 2006. The direct3d 10 system. In SIGGRAPH ’06: ACM SIGGRAPH 2006 Papers, ACM Press, New York, NY, USA, 724–734. FRISKEN, S. F., PERRY, R. N., ROCKWOOD, A. P., AND JONES, T. R. 2000. Adaptively sampled distance fields: a general representation of shape for computer graphics. In SIGGRAPH ’00: Proceedings of the 27th annual conference on Com-puter graphics and interactive techniques, ACM Press/Addison- Wesley Publishing Co., New York, NY, USA, 249–254. LOOP, C., AND BLINN, J. 2005. Resolution independent curve rendering using programmable graphics hardware. In SIG-f l o a t di s tAlphaMa sk = b a s eCo l o r . a ; i f ( OUTLINE ( di s tAlphaMa sk = OUTLINE MIN VALUE0 ) ( di s tAlphaMa sk = OUTLINE MAX VALUE1 ) ) { f l o a t o F a c t o r = 1 . 0 ; i f ( di s tAlphaMa sk = OUTLINE MIN VALUE1 ) { o F a c t o r = smo o t h s t e p ( OUTLINE MIN VALUE0 , OUTLINE MIN VALUE1 , di s tAlphaMa sk ) ; } e l s e { o F a c t o r = smo o t h s t e p ( OUTLINE MAX VALUE1 , OUTLINE MAX VALUE0 , di s tAlphaMa sk ) ; } b a s eCo l o r = l e r p ( ba s eColor , OUTLINE COLOR, o F a c t o r ) ; } i f ( SOFT EDGES ) { b a s eCo l o r . a = smo o t h s t e p ( SOFT EDGE MIN, SOFT EDGE MAX, di s tAlphaMa sk ) ; } e l s e { b a s eCo l o r . a = di s tAlphaMa sk = 0 . 5 ; } i f ( OUTER GLOW ) { f l o a t 4 glowTexe l = tex2D ( Ba s eTextur eSampl e r , i . baseTexCoord . xy+GLOW UV OFFSET ) ; f l o a t 4 glowc = OUTER GLOW COLOR smo o t h s t e p ( OUTER GLOW MIN DVALUE, OUTER GLOW MAX DVALUE, glowTexe l . a ) ; b a s eCo l o r = l e r p ( glowc , ba s eColor , mskUsed ) ; } Figure 9: HLSL source code for outline, glow/dropshadow, and edge softness. GRAPH ’05: ACM SIGGRAPH 2005 Papers, ACM Press, New York, NY, USA, 1000–1009. MITCHELL, J., MCTAGGART, G., AND GREEN, C. 2006. Shading in valve’s source engine. In SIGGRAPH ’06: ACM SIGGRAPH 2006 Courses, ACM Press, New York, NY, USA, 129–142. QIN, Z., MCCOOL, M. D., AND KAPLAN, C. S. 2006. Real-time texture-mapped vector glyphs. In I3D ’06: Proceedings of the 2006 symposium on Interactive 3D graphics and games, ACM Press, New York, NY, USA, 125–132. RAY, N., NEIGER, T., CAVIN, X., AND LEVY, B. 2005. Vector texture maps. In Tech Report. SEN, P. 2004. Silhouette maps for improved texture magnifica-tion. In HWWS ’04: Proceedings of the ACM SIGGRAPH/EU-ROGRAPHICS conference on Graphics hardware, ACM Press, New York, NY, USA, 65–73. TUMBLIN, J., AND CHOUDHURY, P. 2004. Bixels: Picture sam-ples with sharp embedded boundaries. In Rendering Techniques, 255–264. c 2007 Valve Corporation. All Rights Reserved
  • 5. REFERENCES REFERENCES Figure 8: Corner encoded at 64x64 using one distance field (left) and the AND of two distance fields (right) c 2007 Valve Corporation. All Rights Reserved