EigenFace Fisherface

13
Eigenfaces vs. Fisherface s

description

Overview of Bellhuemer paper on Eigen Face and Fisher Face.

Transcript of EigenFace Fisherface

Page 1: EigenFace Fisherface

Eigenfaces vs. Fisherfaces

By

Abdul Malik Khan

Page 2: EigenFace Fisherface

INTRODUCTIONFacial recognition is a problem of automatically identifying or verifying a person from a digital image

For the given paper the problem is: Given a set of face images labeled with the person’s identity (the learning set) and an unlabeled set of face images from the same group of people (the test set), identify each person in the test images.

MethodsIn this paper, following four pattern classification techniques for solving the face recognition problem are studied

CorrelationEigenface Linear Subspace Fisherface

Face images databaseTo implement the paper, yale cropped database is downloaded3

Database contains varied light source location of 38 persons, database is divided into 5 subsetsSubset 1: Light source is 0~15° off camera axis [6 images]Subset 2: Light source is 15~30° off camera axis [10 images]Subset 3: Light source is 30~45° off camera axis [4 images]Subset 4: Light source is 45~60° off camera axis [7 images]Subset 5: Light source is 60~75° off camera axis [4 images]

The data-base images are given in appendix-A

EXPERIMENTSVariation in Lighting

Extrapolation Interpolation

Variation in Facial Expression, Eye Wear, and LightingFull face Closely cropped face

Not Done due to time constraints

Page 3: EigenFace Fisherface

ExtrapolationEach method was trained on samples from Subset 11 and then tested using samples from Subsets 1, 2, and 32. Since there are 190 images in the training set, correlation is equivalent to the Eigenface method using 189 principal components. Next Figure shows the result from this experiment.

Extrapolation Results

1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 30

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Subset 1 - 2 - 3

Err

or

Rate

Extrpolation Results

Correlation

Eigen Face - 4

Eigen Face - 10

Eigen Face - 4 (wo-3)

Eigen Face - 10 (wo-3)

Fisher Face - 10 (wo-3)

25 PCA Weights

1 Subset 1 contains 228 images for which both the longitudinal and latitudinal angles of light source direction are within 15° of the camera axis, 38 images are used as test sets

2 To test the methods with an image from Subset 1, that image was removed from the training set, i.e. the “leaving-one-out” strategy was employed

Page 4: EigenFace Fisherface

25 Fisher Weights

Eigen Values magnitude

0 50 100 150 200 25010

-15

10-10

10-5

100

105

Eigen Values

Mag

nitu

de (

dB)

Eigen values magnitude for Eigen Face

Page 5: EigenFace Fisherface

0 20 40 60 80 100 120 140 160 180 20010

-20

10-15

10-10

10-5

100

105

1010

Eigen Values

Mag

nitu

de (

dB)

Eigen values magnitude for Fisher Discriminator, Negetive values are in Red

Page 6: EigenFace Fisherface

InterpolationEach method was trained on samples from Subset 1 & 53 and then tested using samples from Subsets 2, 3 and 4. Next Figure shows the result from this experiment.

25 PCA Weights

25 Fisher Weights

3 Subset 1 contains 228 images for which both the longitudinal and latitudinal angles of light source direction are within 15° of the camera axis, Subset 5 contains 152 images for which effective longitudinal and latitudinal angles of light source direction are between 60 and75° of the camera axis

Page 7: EigenFace Fisherface

Interpolation Results

2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 40

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Group 2 - 3 - 4

Err

or R

ate

Interpolation Results

CorrelationEigen Face - 4

Eigen Face - 10

Eigen Face - 4 (wo-3)

Eigen Face - 10 (wo-3)

Fisher Face - 10 (wo-3)Fisher Face - 25 (wo-3)

Eigen Values magnitude

Page 8: EigenFace Fisherface

0 50 100 150 200 250 300 350 40010

-14

10-12

10-10

10-8

10-6

10-4

10-2

100

102

104

Eigen Values

Mag

nitu

de (

dB)

Eigen values magnitude

0 50 100 150 200 250 300 35010

-20

10-15

10-10

10-5

100

105

1010

Eigen Values

Mag

nitu

de (

dB)

Eigen values magnitude for Fisher Discriminator, Negetive values are in Red

Comments on results

References

M. Turk and A. Pentland, “Eigenfaces for Recognition,” J. Cognitive Neuroscience, vol. 3, no. 1, 1991.http://www.cs.princeton.edu/~cdecoro/eigenfaces/http://vision.ucsd.edu/extyaleb/CroppedYaleBZip/CroppedYale.zip

Page 9: EigenFace Fisherface

Group-1

Group – 2

Page 10: EigenFace Fisherface

Group – 3

Group – 4

Group – 5