+ sridhar godavarthy july 01, 2010 defense of a masters thesis computer science and engineering...

53
+ Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida Microexpression Spotting in Video Using Optical Strain

Upload: rodney-elkin

Post on 15-Dec-2015

215 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

Sridhar GodavarthyJuly 01, 2010

Defense of a Masters ThesisComputer Science and EngineeringUniversity of South Florida

Microexpression Spotting in Video Using Optical Strain

Page 2: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

Microexpression Spotting in Video Using

Optical StrainSridhar Godavarthy

Examining Committee

Dmitry B. Goldgof, Ph.D. – Major Professor

Sudeep Sarkar, Ph.D. Rangachar Kasturi, Ph.D.

July 01, 2010

Defense of a Masters ThesisComputer Science and EngineeringUniversity of South Florida

Page 3: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Minutes of the presentationMicroexpressions - “micro” expressions.

Goal: Detect “interesting” sequences containing μE.

Approach: optical flow + strain thresholding.

Result: True positive detection as high as 80%.Good performance on real time videos.

Conclusion: Novel system. Scope for improvement. Need more datasets.

Page 4: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

Introduction

Page 5: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Expressions

Social emotion conveyance

Non verbal

Voluntary or involuntary

6 primary expressions

5

Anger Disgust Fear Happiness Sadness Surprise

Page 6: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Microexpressions – What?

Subtle movements of the human body Observable Insufficient to convey emotion

Masking an expression

1/25th to 1/5th of a second

Almost impossible to suppress

07/01/2010

6

Page 7: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Why?

Lie Detection

Pain detection for autistic and anaesthetized patients

Social signal processing( boredom/ concentration detection)

Psychological counseling.

07/01/2010

7

Page 8: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ State of the art

07/01/2010

8

Microexpression Research

Psychology Vision

Intent FACS classificationOptical FlowGabor FiltersANNsRule Based

Page 9: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Objective

Design a preprocessing system that

Spots microexpressions.

Handles small translational and rotational

motion

Improves performance of existing systems

Greater weight to true positives.

07/01/2010

9

Page 10: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Some Fundamentals

Optic Flow : Vector representation of temporal changes

Strain: Relative deformation of material (skin)

Haar Classifier / Viola-Jones Face detector Cascade of weak classifiers OpenCV implementation Uses Haar rectangular features

Page 11: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Brief overview of algorithm ( Main Idea)Skin deforms during an expression.

Deformation peaks at peak of expression

Duration of increased strain corresponds to microexpression

07/01/2010

11

~22 frames

~5 frames

Peak DetectionThresholding

Frames

Str

ain

Mag

nitu

de

Macro Expression

Micro Expression

Page 12: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

Algorithm

Page 13: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+System Flow

Split Frames Face Detection & Alignment Optical Flow

Strain MapSplit into regionsStrain patterns and

Thresholding

Page 14: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Face detection

Viola-Jones face detector-OpenCV implementation

Page 15: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Face Alignment: Rotation

07/01/2010

15

Page 16: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+System Flow

Split Frames Face Detection & Alignment Optical Flow

Strain MapSplit into regionsStrain patterns and

Thresholding

Page 17: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Optical flow

07/01/2010

17

• Black and Anandan• Dense

MJ Black’s Matlab imlementation of OF

Page 18: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+System Flow

Split Frames Face Detection & Alignment Optical Flow

Strain MapSplit into regionsStrain patterns and

Thresholding

Page 19: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Facial strain

07/01/2010

19

Page 20: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+System Flow

Split Frames Face Detection & Alignment Optical Flow

Strain MapSplit into regionsStrain patterns and

Thresholding

Page 21: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Region Splitting

forehead(fh)

right cheek (rc)

left cheek (lc)

right mouth(rm)

left mouth(lm)

right eye(re)

below mouth(bm)

AUs not covered• Blink• Close eyes• Neck tightening• Nostril flare

Automated. Manual intervention if classifier fails

Page 22: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

Datasets & Results

Page 23: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Datasets

USF: IRB Canal9: EULA Found Videos: Fair use act

USF(100)

Canal9 (24) Found (4)

Page 24: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Threshold Determination

Sl. No.Threshold as percentage of

peak strain% True positives % False positives

175 31.8 0

250 50 0

335 77.2 22.7

430 54.5 36.4

525 13.6 0

07/01/2010

24

75 50 35 30 250

10

20

30

40

50

60

70

80

90

100

Threshold Selection

% Peak Strain

% T

rue

Po

sit

ive

Page 25: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Thresholded Strain Maps – Sample 1 / 3

07/01/2010

25

Page 26: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Thresholded Strain Maps – Sample 2 / 3

07/01/2010

26

Page 27: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Thresholded Strain Maps – Sample 3 / 3

07/01/2010

27

False Positive ( Indicative only)

Page 28: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Microexpression Spotting

07/01/2010

29

Page 29: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Negative Test Case – Rejects Expressions

07/01/2010

30

Page 30: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

Concluding Remarks

Page 31: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Contributions and Conclusions

Automated thresholding

Automated alignment (Partial)

Region wise detection

Up to 80% true detection

Microexpressions with expressions are detected.

Page 32: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Constraints

Constant illumination

Neutral face

Some expressions may be falsely detected

Talking

Page 33: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Future Work

Dataset CollectionReal time questioning videos

Fully automated face alignmentBy matching optical flow vectors

Automatic identification of neutral face

Automatic portioning of facesAnthropomorphic landmark identification

07/01/2010

34

Page 34: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Related Publications

Shreve, M., Godavarthy, S., Manohar, V., Goldgof, D., Sarkar, S., "Towards macro- and micro-expression spotting in video using strain patterns," Workshop on Applications of Computer Vision, 2009 pp:1-6

07/01/2010

35

Page 35: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Other Publications by author Candamo, J., Kasturi, R., Goldgof, D., Godavarthy, S., "Detecting Wires in

Cluttered Urban Scenes Using a Gaussian Model, " to appear in Proceedings of International Conference on Pattern Recognition(ICPR 2010), Turkey, 2010

Godavarthy, S., Roomi, M. Md., “Adaptive Contrast Based Unsharp Masking,” in Proceedings of the National Workshop on Computer Vision, Graphics and Image Processing, Feb 2002

Godavarthy, S., Pandian, A., Roomi, M. Md., “Histogram Equalization by Measure of Enhancement,” in Proceedings of the National Workshop on Computer Vision, Graphics and Image Processing, Feb 2002

Godavarthy, S., Shankar, A., Roomi, M. Md., “Adaptive Watermarking-a FFT Approach”, Proceedings of International Conference on Advances in Telecommunication and Information Technology "Asia - Pacific Telecom 2000" (14th, 15th December 2000), Vellore

07/01/2010

36

Page 36: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Dr. Ekman on A-Rod

http://www.nytimes.com/2009/02/15/weekinreview/15marsh.html

Page 37: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

THANK YOU

38

Page 38: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Index Presentation:

Minutes

System Flow

Thresholding

Sample Strain Maps

Results

Negative Test Case

Conclusion and Future Work

Additional Slides

Evolutionary psychology

Detailed Flow Chart

Optical Flow

Elasticity and Strain

FACS

OF Vs OS

Dataset Details

Page 39: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

Additional Slides

Page 40: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

Study of everything we discussed until now

The child of ONE man - Paul Ekman. Over thirty years of research One of the world’s leading experts on lying. About 2 dozen books and innumerable articles Developed FACS Scientific Advisor to “Lie to Me” Co creator of Microexpression Training Tool (METTx)

Evolutionary Psychology

Page 41: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Flow Chart

07/01/2010

42

Split into individual frames

Detect and Crop face

Translate/Rotate to align with neutral mage

Calculate Robust Optical Flow

Calculate Optical Strain

Displacement Vectors

Split into Regions

Compute and normalize strain/region

Strain per pixel

Eight regions or less depending on visibility

Thresholding for magnitude, duration and spatial locality

Microexpression

Input video sequence

Viola-Jones Face Detector

Haar Cascade, Skin Detection

Finite Difference Method

Black and Anandan Method

Determine threshold for Strain Magnitude

Threshold

Page 42: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Add: OFMotion Estimation: Optical Flow Method

Reflects the changes in the image due to motion

Computation is based on the following assumptions: observed brightness of any object point is constant over time nearby points in the image plane move in a similar manner

Minimization problem: (brightness const.) (smoothness const.)

Robust estimation framework (Black and Anandan, 1996) Recast the least squared formulations with a different error-norm function instead of

quadratic Coarse-to-fine strategy

Construct a pyramid of spatially filtered and sub-sampled images Compute flow values at lowest resolution and project to next level in the

pyramid

43

)()(),( 222222yxyxtyx vvuufvfufyxE

10/29/2009

Page 43: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

• Def: Optical Flow is the apparent motion of brightness patterns in the image

• Ideally, same as the motion field

• Have to be careful: apparent motion can be caused by lighting changes without any actual motion

Optical Flow

• Key assumptions• Brightness constancy: projection of the same

point looks the same in every frame• Small motion: points do not move very far• Spatial coherence: points move like their

neighbors

Page 44: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

Elasticity

Different materials have different elasticity

Elasticity can be modeled

strain

stressElasticity Known

Calculate

Page 45: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

What is Facial Strain? Strain on soft tissue when expressions are made. Anatomical method Uses a pair of frames to measure deformation

Facial Strain

Page 46: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+

Finite Difference Method

Compute spatial derivatives from discrete points. Forward Difference Method Central Difference Method Richardson extrapolation

Strain Measurement

Page 47: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Thresholding

Threshold Strain Maps to segment out μE

07/01/2010

48

Page 48: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+The Facial Action Coding System (FACS)

Coding of human expressions

Observational and Anatomical

32 Action Units and 14 Action Descriptors

Encode any possible [facial] expression.

Also used for facial expression simulations

Page 49: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Name Method Type Content

Advantages Disadvantages

FAST O T S Early method Only negative emotions

FACS O A C All muscles Allows for

discovery

-

MAX O T S Faster performance

Only pre defined configurations.

EMG Obt A C Muscular activity invisible to naked eye

Interference from nearby muscles

EMFACS O T S Faster performance.

Only certain emotion expressions.

O - Observational Obt - Obtrusive A – Anatomical T – Theoretical S – Selective C – Comprehensive

Page 50: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+FACS examples

07/01/2010

51

Page 51: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Why Optical Strain?

07/01/2010

52

Page 52: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+ Datasets

Dataset Name

No. of Sequence

s

Approximate Duration

per sequence(s)

Microexpressions

per sequenc

e

Total Resolution

USF – feigned 12 140 8 96 SD / HD

USF – questioning 4 65 1 4 HD

Canal9 dataset 6 300-400 4 24 HD

Found videos 3 30-40 1 4 Very Low

07/01/2010

53

Page 53: + Sridhar Godavarthy July 01, 2010 Defense of a Masters Thesis Computer Science and Engineering University of South Florida

+Index Presentation:

Minutes

System Flow

Thresholding

Sample Strain Maps

Results

Negative Test Case

Conclusion and Future Work

Additional Slides

Evolutionary psychology

Detailed Flow Chart

Optical Flow

Elasticity and Strain

FACS

OF Vs OS

Dataset Details