sunee holland university of south australia school of computer and information science supervisor:...
TRANSCRIPT
Evaluation of a Facial Expression Technique Portraying Deception
Sunee Holland
University of South AustraliaSchool of Computer and Information Science
Supervisor: Dr G Stewart Von Itzstein
Presentation Overview
What is Facial Expression Synthesis? Background Motivation Methodology
› Research Questions Implementation
› Why the Source SDK? › System Architecture› Script Design› Voice Recording› Source SDK – Face Poser› Source SDK – Hammer
User Study Results Conclusion
What is Facial Expression Synthesis?
Simulation of human facial communication into graphical form
The goal is to increase realism and enhance the user’s experience
Can extend facial expression synthesis techniques to show deception leakage in virtual faces
Background
Techniques for facial expression synthesis› Motion capture› Muscle based actuation› Deformations› Morphing or blending
Toolkits› Source SDK (Face Poser), Xface, BEAT, Face
Toolkit, Expression
Background
Embodied Conversational Agents (ECAs)› Realistic, virtual avatars› Gestures› Facial expressions› Speech
Psychology› Deceptive expressions in humans› Paul Ekman’s universal emotions› Micro expressions› Duchenne de Boulogne
Motivation
The aim of this research is to evaluate the means of communication between human and computer› Specifically, deception in virtual agents› Using micro expressions
Literature covers how facial expression synthesis is performed
We want to focus on what these systems can be used for
Methodology
Question the feasibility of deceptive facial expression synthesis
Develop a deceptive facial expression synthesis technique
Evaluate this technique through a user study
Research Questions
Is it feasible to create a facial expression synthesis technique that portrays deception?› Is the user able to detect and recognise
deception leakage on a computer generated model?
› What effect does the synthesised deception leakage of an animated 3D character have on the user's experience?
Research Questions
What are the challenges in implementing a technique for synthesising deception leakage in computer-generated characters?› What existing software is available for
synthesising facial deception leakage and what is the optimal choice for portraying deception?
› How can deception be implemented in facial expression synthesis with the tools currently available?
Implementation
Using the Source SDK Design the scenarios used in the
experiment Generate audio dialogue to create
choreographies in Face Poser Use those choreographies to make
maps in Hammer Map runs through the Source Engine in
Half Life 2
Why the Source SDK?
Other toolkits evaluated: › Xface, rFace, Expression
Need high level of detail so that subtlety can be accurately portrayed
Script Design
Need to create a script for the scenarios to be included in the experiment
As we want to evaluate the expressions only, must remove other confounding factors
Contextual bias could be a confounding factor› Bad: “I can’t wait to see your parents. It will be great”› Good: “I can’t wait to see my parents. It will be great”
Need to express emotions equally› Anger, disgust, fear, happiness, sadness and surprise,
Duchenne (genuine) smile and Pan American (non-genuine) smile
Scenarios need to be realistic
Voice Recording
Choosing voice actors› One male and one female
They were given a script and asked to voice the dialogue› Recorded using computer software› Encoded into a format Face Poser works with› WAV 16-bit PCM
The speech was required to be said in a neutral tone as to avoid bias induced in the voice
Source SDK – Face Poser
SDK tool that produces choreographed sequences
Audio files are imported› Phoneme extraction/lip synching
Choreographies are created using a timeline of events› Contains expressions and audio
Result is exported into Valve Choreography Data (VCD) file
Source SDK – Face Poser
Source SDK - Hammer
SDK tool to create, edit, and export maps for Source Engine games
Build the physical geometry Place entities in the map
› Player spawn location, triggers, map changes, choreographies, characters
Place an event on a generic character that links to a choreography file
Compile and run through Half Life 2
Source SDK - Hammer
System Architecture
User Study
In the process of finalising the experiment Participants
› Ideal population: over 30 participants› Students, researchers and staff
Training phase› Some people will have a predisposition to
recognising deception› Reduce this confounding factor by placing all
participants on a similar experience level
User Study
1. Evaluating the recognition of expressions› How well does the technique express
emotion?› User is shown an expression or scenario
and asked whether they thought it was genuine or deceptive, and/or what emotion they thought they were perceiving
User Study
2. Evaluating subjective user experience› What level of quality is the interaction
between computer and user?› Participants will be asked to fill out a
feedback form › Subjective feedback wrt how easily they
could identify deception and emotions
Results
Predict that there will be difficulty in accurately recognising deception› Similar with humans
However, it should show a recognition rate that is slightly greater than chance› This is all that is required to validate the
research
Conclusion
The main contribution of this research is the evaluation of facial expression synthesis portraying deception
The user study will answer the question regarding feasibility through recognition rate and subjective user experience
Question regarding challenges is answered in the implementation through selecting a pre existing solution and describing how the deception was implemented
Future Work
More complex facial expressions More ways of portraying deception
other than facially› Body leakage is important when detecting
deception Possible applications in gaming to
improve the immersive experience
Questions?
Experimental Software
Want a way to randomise the choreography files for every participant in the experiment
Solution: Custom software that inputs files from a location, randomises the ordering, then outputs in the Half Life 2 scenes directory
Written in C#