Cloning Facial Expressions with Userdefined Example Models

1
Cloning Facial Expressions with User-defined
Example Models
2
Introduction

• High-quality facial animations
• Simulating physical behaviors
• Surface deformation
• Reusing existing motion data

Motion Data
3
Related Work

• Traditional Facial Approach
• Physical-based approach Waters 95, Lee 95
• Feature point-driven approach Pighin 98,
  Guenter 98
• Performance-driven approach Williams, 90
• Facial Retargeting Approach
• Expression Cloning Noh, 01
• Example-based approach
• Motion Capturing Cartoons Bregler, 02
• Scattered Data Interpolation Rose, 98, Sloan,
  01

4
Motivation

• Expression Cloning Noh, 98
• Characteristic features of a new model
• Artists intention in the output animation
• Motion vector calculation
• ? Example-based Cloning Approach

5
Overview
6
Key-Model Construction (1)
??
?
7
Key-Model Construction (2)

• Importance-based approach
• Phoneme-important region
• Expression-important region

High phoneme-importance
Large movements for pronunciation
High expression-importance
Small movements for pronunciation
8
Key-Model Construction (3)
1st phoneme-importance expression-importance
2nd phoneme-importance expression-importance
final phoneme-importance expression-importance
9
Key-Model Construction (4)

• Composition based on importance

Movement for speech
v
v
Movement for expression
For the phoneme-important region
10
Parameterization
Parameters
11
Parameterization
Source Base Model SB
?Vi (p1, p2, , pn)
ith Source Example Model Si
12
Parameterization
Si
?Vi
Ti
Parameter Space
Example Models
SB
13
Expression Blending
Parameters
14
Expression Blending
15
Predefining Weight Function

• Multidimensional Scattered Data Interpolation
  Sloan et al, 00
• Linear Approximation Radial Basis Interpolation
• For each target example, Ti, compute a smooth and
  continuous weight function

Weight for T1
1
Parameter
T1
T2
T3
T4
16
Result Video
17
Conclusion

• Example-based cloning approach
• Characteristic features of a new model
• Artists intentions in the resulting animation

18
Future Work

• Blending face model regionally
• Partition face model into upper and lower part
• Diverse expressions with a smaller number of
  example models
• Automatic search for feature points
• Eye blinking, hair, tongue animation