Facial Expression Recognition-Implementation and Applications In Computer Sciences-Project Presentation, Slides for Applications of Computer Sciences. Bengal Engineering & Science University
padmavati
padmavati18 July 2012

Facial Expression Recognition-Implementation and Applications In Computer Sciences-Project Presentation, Slides for Applications of Computer Sciences. Bengal Engineering & Science University

PDF (523.4 KB)
27 pages
674Number of visits
Description
This presentation is for final year project to complete degree in Computer Science. It emphasis on Applications of Computer Sciences. It was supervised by Ambalika Virochan at Bengal Engineering and Science University. I...
20points
Download points needed to download
this document
Download the document
Preview3 pages / 27
PowerPoint Presentation

Facial Expression Recognition: An Overview

 Definition:

 Facial expression recognition deals with the classification of facial features into classes based on visual information.

 Guessing the meaning of facial deformations

 Three phases of recognition:

 Face Acquisition

 Facial Feature Extraction

 Expression Classification

docsity.com

Facial Expression Analysis framework

 Most of the facial expression recognition systems

make use a simple facial expression analysis

framework

docsity.com

Facial Expression Recognition Techniques

 Researchers in the recent past have been

trying to automate this task on a computer.

 Approaches for facial expression analysis

from both static images and video have been

proposed in the literature.

 Researchers are using a combination of

image/ video processing techniques, along

with machine learning techniques like artificial

Neural Networks.

docsity.com

Facial Expression Recognition Techniques

 Feature Extraction Techniques:  PCA

 Wavelets

 Feature Point Tracking

 Optical Flow

 Expression Classification Techniques:  Learning Vector Quantization (LVQ).

 Support Vector Machines (SVM)

 Hidden Markov Models (HMM)

 Feed-Forward Back Propagation Network

docsity.com

Facial Expression Recognition Using PCA

 PCA

 Principle Component Analysis

 A technique useful for compression and classification of data

 Purpose

 To reduce the dimensionality of the dataset

 Finding new set of variables, smaller than the original set of variables, that contain most of the sample’s information.

docsity.com

Facial Expression Recognition Using PCA

 Step 1: Preprocessing

 Input images are of size 256x256

 Resize images into 140x100

 Reshape 2D image into a 1D vector

 Combine all 1D vectors into a Matrix i.e. X

docsity.com

Facial Expression Recognition Using PCA

 Step 2: Feature Extraction

 Calculate the Mean of the Matrix X.

 Subtract Mean from All Vectors in X. i.e.

 A = X - Mean(X)

 Get the covariance matrix

 the covariance matrix is a matrix of covariance between

elements of a vector.

 L = A'*A;

 Get the eigenvectors and eigenvalues

 [Vectors,Values] = eig(L);

docsity.com

Facial Expression Recognition Using PCA

 Step 3: Expression Classification

 For classification i used LVQ.

 Create a learning vector quantization network.

 Goal is to have the network "discover" structure in the

data by finding how the data is clustered.

 In Matlab, an LVQ network can be created with the

function newlvq

 We require input vectors and target expression classes

to train the neural network.

docsity.com

Training Using LVQ Network

Training Using LVQ Network

No. Of Training Samples 191

No. Of Testing Samples 22

Sample Size 140x100

Expression Labels Happy

Fear

Sad

Surprise

Disgust

Anger

Natural

No. of Epochs 150

 Leave One subject out

 Total 213 images

 10 subjects

 Training Data

 9 subjects

 191 images

 Testing Data

 1 Subject

 22 images

docsity.com

Training Results

Training Results

No. of

Misclassificati

ons

36/191

Recognition Rate 80.15%

Error Rate 18.85%

docsity.com

Confusion Matrix of Training Result

Anger Disgust Fear Happy Natural Sad Surprise

Anger 25/27

=92.6%

0% 0% 0% 0% 2/27 =7.41% 0%

Disgust 4/26

=15.4%

19/26

=73.08%

2/26

=7.69%

0% 1/26

=3.85%

0% 0%

Fear 0% 0% 22/28

=78.57%

0% 2/28

=7.14%

4/28

=14.29%

0%

Happy 1/28

=3.57%

0% 1/28

=3.57%

25/28

=89.29%

1/28

=3.57%

0% 0%

Natural 0% 0% 1/27

=3.70%

3/27

=11.11%

18/27

=66.67%

5/27

=18.52%

0%

Sad 4/28

=14.3%

0% 0% 1/28

=3.57%

2/28

=7.14%

21/28

=75%

0%

Surprise 0% 0% 1/27

=3.7%

0% 0% 0% 26/27

=96.3%

docsity.com

Testing Results

Testing Results

No. of Misclassifications 7/22

Recognition Rate 68.18%

Error Rate 31.82%

docsity.com

Result Improvement

 Two methods are used

 By changing the range and number of Principle

Components

 By Extracting Mouth and Eyes and then apply

Principle Component Analysis

docsity.com

Result Improvement I

 PCA is also used for face recognition

applications.

 The Principle components that have higher

Eigen values may or may not correspond to

varying expression.

 i have used first 50 Principle Components

 But in order to improve the results i have tried

different ranges of principle components

docsity.com

Result Improvement Using 30-60

Principle Components

Table 7: Recognition Using 30-60

PC's and 150 Epochs

No. of

Misclassifications

14/213

Recognition Rate 93.43%

Error Rate 6.57%

docsity.com

Result Improvement Using 20-50

Principle Components

Recognition Using 20-50 PC's and

500 Epochs

No. of

Misclassifications

10/213

Recognition Rate 95.31%

Error Rate 4.69%

docsity.com

Result Improvement II

 For improvement, following steps are used

 Extract Both eyes from image

 Extract mouth from image

 Reshape images into vectors

 Combine all vectors in form of Matrix

 Apply PCA on Matrix for Feature extraction

 Apply LVQ for Feature Classification

docsity.com

New Framework

docsity.com

Training Results

Recognition Using 20-50

PC's and 150 Epochs

No. of

Misclassificatio

ns

11/191

Recognition

Rate

94.24%

Error Rate 5.76%

docsity.com

Testing Results

Testing Results

No. of Misclassifications 10/22

Recognition Rate 54.54%

Error Rate 45.46%

docsity.com

Facial Expression Recognition Using 2D

Gabor Wavelets

 Gabor Filter

 Linear Filter.

 Its impulse response is defined by the Harmonic

Function multiplied by the Gaussian Function.

docsity.com

Basic Framework for Gabor Filtering

docsity.com

Results Using 2D Gabor Filtering

 Following results are generated using 1-50

Principle Components

Results Using Gabor Filter

No. of

Misclassificatio

ns

29/191

Recognition Rate 84.82%

Error Rate 15.18%

docsity.com

Result Improvement

 Following results are generated using 20-50 Principle

Components

Improved Results Using

Gabor Filter

No. of

Misclassificati

ons

21/191

Recognition Rate 89.01%

Error Rate 10.99%

docsity.com

Time Schedule

ID Task Name Start Finish Duration

2006 2007

Jun Jul Aug Sep

1 3w6/6/20065/17/2006Project Proposal

2 5w7/11/20066/7/2006SRS Document

3 3w7/12/20066/22/2006Project Plan

4 6w8/22/20067/12/2006Literature Survey

5 2w9/5/20068/23/2006Technique Selection

6 5w10/9/20069/5/2006Feature Extraction

7 5w11/10/200610/9/2006Feature Classification

8 5w12/14/200611/10/2006Technique Implementation

9 10w2/21/200712/14/2006Results Improvement

10 8w4/18/20072/22/2007Software Implementstion

11 8w6/13/20074/19/2007Thesis Writing

Oct Nov Dec Jan Feb Mar Apr

docsity.com

Future Work

 Up till now, I have implemented

 Feature Extraction Techniques:  PCA

 Wavelets

 Expression Classification Techniques:  Learning Vector Quantization (LVQ).

 Currently I am working on

 Expression Classification Technique  Support Vector Machines (SVM)

docsity.com

Thanks

Questions?

docsity.com

comments (0)
no comments were posted
be the one to write the first!
This is only a preview
See and download the full document
Docsity is not optimized for the browser you're using. In order to have a better experience please switch to Google Chrome, Firefox, Internet Explorer 9+ or Safari! Download Google Chrome