On Identification of Face Invariance

1
On Identification of Face Invariance

• Devi Vijayan
• M.Tech Computer Vision Image Processing
• Amrita School of Engineering

2
Outline

• Motivation
• Objective
• Approach
• Future Work

3
Intra - class Variability

• Faces with intra-subject variations in pose,
  illumination, expression, accessories, color,
  occlusions, and brightness.

4
Inter - class similarity

• Different persons may have very similar
  appearance.

5
Objective To identify the invariant facial
factors
Hypothesis

• No single visible feature such as eyes, ear or
  nose can identify though they make vital
  contribution to identification of face
  invariance.
• Lower order combinations of these features
  termed as Face Invariant Factors (FIF).

6
Approach

• A hybrid model that combines the properties of
  Principal Component Analysis (PCA) and Geometric
  Feature based Techniques.
• Randomized higher-order eigenfaces based model.

7

Principal Component Analysis (PCA)

1. Acquire a set of M fixed sized face images,
known as the training set.

N X N
N2 x 1

.
N2 X 10
1 2 ..10
8
PCA 1

• 2 Find the average face mof the face set.
• ? 1/M (?n Gn)
  n 1,2..M
• 3 Find the deviation of each face image from the
  mean ,
• F i G i -? i 1,2 M
  

Averaging
. .
9
PCA 2

• 4. Find the Covariance matrix C.
• C 1 / M ?i ?i ? i T
  i 1,2 M
• Another way of representing the covariance matrix
  is by writing
• A F 1, F 2......... F M
• C 1 / M ? i A AT

X
N2 X 10
10 X N2
N2 x N2
10
PCA 3

• If the number of data points in the image space
  is less than the dimension of the space (M lt N2),
  there will be only M-1, rather than N2,
  meaningful eigenvectors.
• Construct a matrix L ATA, of M x M dimension
  ,and find M eigenvectors
• Corresponding eigenvalues allow us to rank the
  eigenvectors according to the significance.

x
10 x 10
N2 x 10
10 x N2
11
PCA 4
Sorting
Find Eigen vectors
10 x 10
Plot of Eigenvectors
12
When a new test image comes,
? Test image

• Project it into face space
• Determine which image provides the best
  description of input image.
• This is done by minimizing the Euclidean
  distance .

13
Experiment -1. Capture the invariance from
different persons. (Dataset
1)
Reconstructed image
Training set- Data set 1
Test image
14
? Test image
Images
Reconstructed image
15
Experiment -2. Capture invariance from same
person under different disguises.
(Dataset-2)
Training Set
Test Image
Converged image
16
Experiment 2 1







Test image

• 
  
• 
  

Converged image
17
Experiment -3Capturing invariance from same
person under different ages
Training Set(2, 5, 8, 28, 29, 43)
Test Set (Age-10)
18
Experiment 3 1
19
Experiment 3 2
Accuracy of convergence is 59. 25
20
Illustration of Geometric based concept
21
Illustration of Geometric based concept 1

• Inference1 Ratio of the distances between lips
  and nose found to be invariant. (Average distance
  ratio1.215).
• Inference2 Ratio of the area of the face to
  the forehead area seems to be more or less
  invariant. (Average area ratio5.21)
• Inference3 The eye-to-eye distance variation
  is not maintaining a constant increment

22
Using higher order pixel information
Representation seems to be better
23
Applications 1

• 1.Simulating Aging Effects

24
Applications 2

• 2.Developing Caricatures
• Caricatures of faces can be obtained by
  exaggerating their distinctive features.

25
Applications 3

• 3.Mug-shot detection

26
References

• 1 Roberto Brunelli and Tomaso Poggio Face
  Recognition Features versus Templates, IEEE
  Transactions on Pattern Analysis and Machine
  Intelligence, Vol. 15, No. 10, October 1993.
• 2 Turk,M., and Pentland.A,Eigenfaces for
  Recognition, J. Cognitive Neuroscience
  3(1),1991, pp. 71-86.
• 3 Peter N.Belhumeur, Joa P Hespanha, and David
  J.Kriegman, Eigenfaces vs Fisherfaces
  Recognition Using Class Specific Linear
  Projection, IEEE Transactions on Pattern
  Analysis and Machine Intelligence, Vol 19, No
  7,July 1997, pp. 711-720.
• 4 A. Lanitis, C. J. Taylor, T. F. Cootes,
  Towards Automatic Simulation of Aging Effects
  on Face images, IEEE Transactions on Pattern
  Analysis and Machine Intelligence, Vol 24, No.4,
  April 2002,pp. 442-455.
• 5 W, Zhao, R. Chellappa, P.J.Philips, A.
  Rosenfeld, Face Recognition A literature
  survey, ACM Computing Surveys, Volume 35, No.
  4,December 2003,pp.399-458.
• 6 Marian Stewart Bartlett, Javier R. Movellan,
  and Terrence J. Sejnowski, Face Recognition by
  Independent Component Analysis IEEE Transactions
  on Neural Networks, Vol. 13, No.6, November2002,
  pp 1450-1464.
• 7 Kun Peng, Liming Chen, A Robust Algorithm
  for Eye Detection on Gray Intensity Face without
  Spectacles, JCST Vol. 5 No. 3 October 2005.