5S-Screen Space Perceptual Rendering of Human Skin

Screen-Space Perceptual Rendering
of Human Skin
Jorge Jimenez, Veronica Sundstedt and Diego Gutiérrez
The problem of skin rendering
BRDF vs. BSSRDF
Index
 What is subsurface scattering and why it is
important
 Texture-space approach
 Our screen-space approach
 Perceptual validation
 Results
 Future work
Diffusion profiles
 A diffusion profile R is a function that defines how the light
attenuates as it travels beneath the surface of an object.
 In homogeneous objects, the diffusion profiles are radically

symmetric, they only depend on r.
[EON07]
Diffusion profiles
DIPOLE (JENSEN01) MULTIPOLE (DONNER05)
Irradiance maps
 Applying a diffusion profile implies calculating the
irradiance at incident and adjacent points, this
implies repeating calculations.
 If we store the incoming light at each point of the

surface in an irradiance map, we can calculate the
lighting at each point once.
 Using this map, applying a diffusion profile consists

in a simple but costly 2D convolution.
Irradiance maps
 Simply stores the incoming lighting at each point of
the surface.
Irradiance maps
Approximating diffusion profiles
with gaussians
Guassians
[EON07]
Texture-space approach problems
 Backface culling and viewport clipping optimizations lost
 Irradiance map size needs to be managed
 Early-Z rejection optimization lost
 Each subject to be rendered requires her own irradiance
map
 The irradiance map forces the transformation of the
model vertices twice
 Lighting is usually calculated twice
 Seams in texture space cause artifacts in screen space
Details in the paper.

Index
important
 Results
 Future work
Texture- vs. screen space
 The main concept of our work is the translation of the
diffusion from texture to screen space
The Big Picture
 The use of the depth map makes our process different

from simply blurring the image in Photoshop
Kernel width modulation
 Pixels need narrower kernels when:
• They are far from the camera
• They representing surfaces that are at steep angles
with respect to the camera
Demo
Index
important
 Results
 Future work
Perceptual Validation
 The realism of the shaders was validated using
a 2AFC preference experiment
 Comparing four different shaders
• No SSS, TS, SS Jittered, and SS Full
Participants
 16 volunteers (14M, 2F; age range: 23-43)
 Naive as to the purpose of the experiment
 Normal or corrected-to-normal vision
 Varied experience
• 7 reported no expertise in art (drawing, photography,
computer graphics, digital art,...)
• 9 reported that they had this expertise
Hypotheses
 No SSS, SS Jittered and SS Full would produce
as high visual realism as the ground-truth TS
shader
 Participants with expertise in CG are better at
judging realism
Stimuli
 XYZRGB head model
 12 images per shader
• 3 camera distances (28, 50, 132)
• 4 light angles (0°, 60°, 110°, 150°)
 72 trials = each comparison twice
Setup
 Images were displayed on a monitor with a
1650 x 988 resolution
 Dimmed lighting
 Participants were seated on an adjustable
chair with their eyelevel ~ level with the
screen centre
 Viewing distance was ~ 60 cm
Procedure
 Consent form and questionnaire
 2AFC (no reference)
 Task description
• Choose the image which you think look most realistic
(i.e. most like real human skin)
• Unlimited viewing time (10 sec recommended)
• Person has the same age (45 years old)
Results
Index
important
 Results
 Future work
Performance
 Against a texture-space implementation
Limitations
 Small haloes produced by incorrect diffusion from nose to
cheek
 We cannot account for the diffusion produced in thin, high
curvature features
Forward Scattering
 Forward scattering should be seen through thin parts (like the ears).
 Current forward scattering algorithms rely on textures-space

diffusion and thus they cannot be easily translated to screen-space
Image taken from Whelan’s Thesis

Future work
Conclusion
Our algorithm, when compared against a texture-
space approach:
 It brings back GPU implicit optimizations

 Scales better as the number of objects increases
 Reduces artist side efforts
 Has a very simple implementation
 Being a post-process it will not require major
changes to be integrated into a 3D engine
Experiment Conclusions
 Our new shader can produce stimuli with as
high visual realism as a previously developed
physically based shader
 No statistical difference in the number of error
selections made by experienced vs. non
experienced participants (p > 0.05)
Acknowledgments
 XYZRGB Inc. (for the high quality scans)
 The members of the GV2 group at Trinity College Dublin
and everyone that participated in the experiments
 Carl Vogel (for his help with revising the consent and
questionnaire forms)
 Spanish Ministry of Science and Technology (TIN2007-
63025)
 Gobierno de Aragón (OTRI 2009/0411)
 I3A (research grant)
 Josh Checa (for his longtime support)
Thanks for you attention

5S-Screen Space Perceptual Rendering of Human Skin

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

5S-Screen Space Perceptual Rendering of Human Skin

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

5S-Screen Space Perceptual Rendering of Human Skin

Uploaded by

Copyright:

Available Formats

Screen-Space Perceptual Rendering

 In homogeneous objects, the diffusion profiles are radically

 If we store the incoming light at each point of the

 Using this map, applying a diffusion profile consists

Details in the paper.

 The use of the depth map makes our process different

 Current forward scattering algorithms rely on textures-space

Image taken from Whelan’s Thesis

 It brings back GPU implicit optimizations

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.