SLIDE 1
Facial expression classification In still images Angel Gutirrez, - - PowerPoint PPT Presentation
Facial expression classification In still images Angel Gutirrez, - - PowerPoint PPT Presentation
Facial expression classification In still images Angel Gutirrez, Montse Pards Introduction Face detection Contour detection Expression estimation System implementation - DEA Results Conclusions Surprise Basic expressions (Ekman i
SLIDE 2
SLIDE 3
INTRODUCTION
· Basic expressions (Ekman i Friesen – 1971)
Joy Sad Neutral Anger Surprise
SLIDE 4
INTRODUCTION
Objective
- To develop an automatic system which can
recognize a facial expression in a still image.
Applications
- Monitor for drivers
- Stress detection
- Facial coder
- Entertainment/ Games
- etc.
SLIDE 5
INTRODUCTION
Generic scheme. Methods.
INPUT IMAGE FACE DETECTION INFORMATION EXTRACTION FACIAL EXPRESSION EXPRESSION DECISION
· Viola and Jones · Region based · Gabor Wavelets filters · Active appearance models · Dense motion fields · Feature points tracking · Hidden Markov Models · Neural Networks · Probabilistic models
SLIDE 6
FACE DETECTION FEATURES DETECTION EXPRESSION DECISION
SLIDE 7
FACE DETECTION
Viola and Jones:
· Fast search · Robust to background · Uses local texture features · Trains classifiers for face/non-face classes · Uses a cascade of classifiers structure (Adaboost)
Region-based:
· Refines the face detection to obtain a better initizialization
SLIDE 8
FACIAL FEATURE CONTOUR DETECTION
Model based method Active Shape Models Active Appearance Models
SLIDE 9
FACIAL FEATURE CONTOUR DETECTION
Based on Active Appearance Model software implemented by Stegmann*
* M. B. Stegmann, B. K. Ersbøll, R. Larsen, FAME - A Flexible Appearance Modelling Environment, IEEE Transactions on Medical Imaging, vol. 22(10),
- pp. 1319-1331, Institute of Electrical and Electronics Engineers (IEEE), 2003
- Facial feature contours are represented by a 58 points model
SLIDE 10
FACIAL EXPRESSION
Class 1, Class 2, ... Class N. Class 1, Class 2, ... Class N.
CLASSIFIER SYSTEM INPUT DATA KNOWN OUTPUT LABELED DATABASE
Bayesian framework with probabilistic model: Mixture of multivariate gaussians, trained with EM for each class
SLIDE 11
RESULTS
Database
· 192 labeled images.
SLIDE 12
RESULTS
SLIDE 13
RESULTS
TEST 1 TEST 1
FACE DETECTION FEATURES DETECTION EXPRESSION DECISION
95,47% H A N Su Sa H 96% 0% 0% 4% 0% A 0% 98% 0% 0% 2% N 0% 0% 94% 6% 0% Su 4% 0% 3% 93% 0% Sa 0% 0% 3% 0% 97%
95,47 %
SLIDE 14
RESULTS
TEST 2 TEST 2
FACE DETECTION FEATURES DETECTION EXPRESSION DECISION
84,66% H A N Su Sa H 96% 0% 2% 2% 0% A 6% 79% 4% 4% 8% N 0% 3% 69% 13% 16% Su 0% 0% 4% 96% 0% Sa 0% 3% 14% 0% 83%
84,66 %
SLIDE 15
CONCLUSIONS
· Automatic system for facial expression detection in 3 stages:
FACE DETECTION FACIAL FEATURE CONTOURS DET. EXPRESSION CLASSIFICATION