Hidden Emotion Detection through Analyzing Facial

Hidden Emotion Detection through Analyzing Facial Expression
Enhao Gong, Tianhe Wang, Jing Xiong
EE 368 Digital Image Processing
Dept. Electrical Engineering, Stanford University
Introduction
Methods
Motivation:
 Facial expression plays an important role in human
communication and interaction.
 Utilizing technology to detect hard-to-notice expressions
smoothens communication process.
 Extracting and highlighting facial features potentially
assists children with autism in studying facial expressions
and emotions.
We developed an Android App using DROID smart
phones/tablets to detect and characterize human facial
expressions. By using image processing and computer vision
algorithms, simple but hard-to-notice facial features can be
highlighted and emotion state can be further analyzed.
1. Feature Detection – Client (Android with OpenCV)
 Implement Viola-Jones face detection and Haar Cascade
classifier algorithm from OpenCV library.
 Sequentially detect faces and eyes & mouths afterwards
 Smart correction and ROI updates
Results
Image Fusion
 Face Images
 Facial feature localization
 Feature & Contour extraction
 Micro-expression Hotspots
Fused Visualization
With AugmentedReality for Emotion
Figure 2: Kernels and method in Viola- Figure 3: Successfully detected face,
Jones detection Haar Cascade Classifier.
eyes, and mouth.
Figure 1.
Snapshot of App.
2. Facial Features & Contour Extraction Server(php,matlab)
 Canny-filter-based extraction
 Corrected ROI-based adaptively adjusting
 Feature-Background-Ratio-based adjusting
Procedure – Flow Chart
Analysis & Application
Figure 6. Image Fusion for
Augmented Reality and Expression
Decoding.
 Faithful feature extraction and highlighting
 Multiple dimensions of information
Future Work
Figure 4: Facial Feature extraction using corrected
ROI&FBR-based adaptively adjusted Canny Filter
3. MicroExpression Hotspot Extraction–Server(php,matlab)
 Covariance & Tracking based Registration
 Characterize motions using
variance in image domain
 Highlighted with ROI based weights
Figure 2: Flowchart of the App.
Decoding: Neutral 70%
Figure 5: Efficient Micro expression
detection using covariance based
registration
and
cross-frame
variance based motion extraction.
Real-time visualization
Further interpretation on micro-expression hotspots
Experiments and training for more accurate hidden
expression decoding
Robust scoring system for hidden emotion
Extend to detect more facial features, gestures and other
emotional states (stress level, lie detection, etc.)
References
1 P. Viola and M. Jones, CVPR 2001
2 G. Bradski, 2000.
3 Essa, Irfan A., and Alex Paul Pentland. 1997
4 Fasel, Beat, and Juergen Luettin, 2003
5 Ekman, Paul, and Wallace V. Friesen. 2003