Facial performance enhancement using dynamic shape space analysis

PRESENTED BY: MOHD IMRAN
Facial Performance Enhancement
using Dynamic Shape Space
Analysis

 Facial Performance Enhancement
 Aims to achieve realistic face modelling
 Fine details like wrinkles
What does the topic mean?

 Facial expression plays a critical role in human interactions
 Human faces can accommodate large range of expressiveness
 Facial expressiveness may differ vastly from person to person
 Grand challenge in the field of computer graphics
Introduction

 3D Face Scanning
 Captures geometry of human faces
 Captures high resolution static poses to be used with marker based motion
capture
 Facial Animation
 Animating images or models of character face
 It can be generated by marker-based/less motion capture
Related Work

 Preprocessing
 Build a dense performance capture database D
 Encode the database into shape space
 Classify into regions
 Spatial Performance Enhancement
 Input animation preprocessing
 Encoding
 Matching
 Interpolation
 Reconstruction
Overview(1)

 Performance Capture Database
 Includes detailed facial geometry
 Acquired using high resolution 3D facial performance capture method
 Passive approach of Beeler et al. [2011] is used
 Data Encoding
 Frequency separation
 Separate low frequency components of D using low pass filter
 Encoding
 For every database frame deformation gradients are encoded and saved
 The low-frequency component are encoded relative to the low frequency neutral pose
 Uniformly cluster the high resolution mesh into patches
 Encode the average deformation gradient for each patch
Preprocessing(1)

 Regions
 Region based face models generalize better than their holistic counterparts
shown by Tena et al. [2011]
 Partition of mesh into groups of common functionality
 Voluntary Regions
 Controlled by the actor directly
 Involuntary Regions
 Areas governed by voluntary regions
Preprocessing(2)

 Input Animation Preprocessing
 Input animations A can come from any source
 Created using a manually controlled rig
 Can be fully controlled manually
 Subset of facial expressions
 Driven by sparse marker-based motion capture
 Registration and Frequency Separation
 Facial properties of face in D and A should be similar
 Obtain a correspondence between database neutral frame and the animation neutral
frame
 Linearly deform the animation A to A1 corresponding to D
 Separate the low frequencies in A1
Performance Enhancement Model(1)

 Encoding
 Same as the encoding in preprocessing
 Matching
 Each region of the face is treated independently
 Mesh triangles have different weights depending on the location
 Each triangle weight in a region is Gaussian function of geodesic distance
 Only the areas within the region participate in this process
 We obtain weights for each frame in database

 Interpolation
 Blends the database within each region according to the weights
 It is a normalized interpolation between regions
 Reconstruction
 Reconstruct the final mesh using Laplace-Beltrami operator

 A framework for data-driven spatial enhancement of low resolution
facial animations, using a compressed shape space
 Validation of the framework on four different types of input facial
animations, with a direct comparison to state-of-the-art
Summary

Facial performance enhancement using dynamic shape space analysis

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Facial performance enhancement using dynamic shape space analysis (20)

Recently uploaded (20)

Facial performance enhancement using dynamic shape space analysis