Title: Seminar: CSE 717 Soft[1] Biometric Traits in Face Recognition System
1Seminar CSE 717Soft1 Biometric Traits in Face
Recognition System
- Problem Description Part
- Zhi Zhang
- zhizhang_at_cse.buffalo.edu 2/21/2004
2Ideal Characteristics of Biometric Traits
- Universality
- Distinctiveness
- Permanence
- Collectability
- Performance
- Acceptability
- Circumvention2
- Regretfully, NONE of the currently using human
biometric traits possesses all of the above
characteristics.
3What is Soft1 Biometric Traits?
- Traditional (Primary) Biometric Traits2
- DNA Sequences
- Iris/Retina
- Fingerprint
- Voice
- Face
- Signature
- The above human biometric traits are relatively
universal, distinctive, permanent and resistant
to circumvent. But they may not be collectable or
acceptable to all the people.
4What is Soft1 Biometric Traits? - Cont
- Soft1 Biometric Traits
- Gender
- Ethnicity
- Eye/Skin/Hair color
- Age
- Height
- Weight
- The above human biometric traits are relatively
LESS distinctive, permanent and resistant to
circumvent. But they provide some evidence about
the user identity that could be exploited1
5Why using Soft Biometric Traits?
- During enrollment, many existing biometric
systems actually collected information like - Gender
- Ethnicity
- Eye/Skin/Hair color
- Age
- Height
- Weight
- If the above traits can be automatically
extracted and incorporated in the decision making
process, the performance of the system can be
improved significantly1.
6Necessary Devices
- Image or Video Device
- As a special Face Recognition System, an image or
video device is a must for both enrollment and
verification/identification. - As color is a relatively important characteristic
for Soft Biometric Traits, the images collected
from the image or video device must be color
images.
7Necessary Devices - Cont
- Auxiliary Devices - Optional
- For those Soft Biometric Traits that can not be
extracted directly from the images, some
auxiliary devices are needed. - If Height trait is expected, an extra height
sensor could be installed to extract this
information. - If Weight trait is expected, an hidden scale
could be installed to extract this information.
8Difficulty Levels of the System
- Verification vs. Identification
- Controlled vs. Uncontrolled
- Database
- Location and Segmentation
- Feature Definition
- Feature Extraction
- Feature Combination
- Matching/Classification
- Decision Making
9Verification vs. Identification
- Verification System
- 1-1 Matching
- Commercially available3
- Identification System
- 1-n Matching
- Still a challenge area
10Controlled vs. Uncontrolled
- Controlled Environment
- Fixed pose
- Simple background
- Special/Fixed illumination
- Uncontrolled Environment
- Free pose
- Complex background
- Different illumination
11Database
- Availability
- FERET4
- Large, 14051 images
- 8-bit greyscale images
- Database from other universities or institutes5
- Variable size
- Color images
- Not standard
- Build our own image database
12Database - Cont
- Selection of the images
- Demographical Distribution
- Gender Distribution
- Age Distribution
- Illumination Distribution - Optional
- Pose Distribution - Optional
- Management of Database
- Indexing
- Binning
13Location and Segmentation
- Behavior-based Agent Model6
- Search the skin-like pixels by a number of
color-sensitive behavior-based agents, which
distributed uniformly in the 2-D image - Mark the face-like region by activating the
evolutionary behavior of the agents - Examine the shape information of each face
candidate region and determine the face region by
fuzzy shape feature analysis - Luminance/Chrominance-Component-based Approach7
- Detect the face location by exploring the
distribution property of the luminance and
chrominance components
14Feature Definition
- Feature definition in Traditional (Primary)
Biometric Traits - Feature definition in Soft Biometric Traits
15Feature Definition in TBT8
- Geometric feature-based method
- Economical representation
- Insensitivity to variations in illumination and
viewpoint - Sensitive to the feature extraction process
- Appearance-based method
- Eigenfaces
- Karhunen-Loeve (KL) Transform or Principal
Component Analysis (PCA) - Most Expressive Features (MEFs)
16Feature Definition in SBT
- Gender Classification Features9
- Feature Selection
- Different eigenvectors encode different kind of
information - Some of the eigenvectors may be irrelevant to
gender classification - Using a Genetic Algorithm (GA) to select a subset
of the eigenvectors - Using the selected subset to train a Neural
Network (NN), which could be applied to perform
gender classification
17Feature Definition in SBT - Cont
- Ethnic Classification Features
- A mixture of experts consisting of ensembles of
radial basis functions for the classification of
gender, ethnic origin, and pose of human faces
was proposed10 - The above work was on FERET database, which means
no color information was utilized - We could acquire the skin color information after
face location and segmentation process - Feature Selection combined with skin color
information, which could be an important feature
in ethnic classification
18Feature Definition in SBT - Cont
- Age Estimation Features
- Relatively a new topic
- A classifier was designed to accept the
model-based representation of unseen images and
produce an estimate of the age of the person in
the image11 - A wrinkle modeling was proposed and a research
about age and gender estimation based on wrinkle
texture and color of facial image was
introduced12 - We could see that both texture and color
information could be applied to age estimation
19Feature Extraction
- A kernel Principal Component Analysis (PCA) was
proposed13 for feature extraction - A nonlinear extension of PCA
- First map the input data into a feature space
via a nonlinear mapping, then apply PCA in the
above feature space - Feature extraction for Soft Biometric Traits
20Feature Combination
- A face verification algorithm based on multiple
feature combination and supporting vector machine
was proposed15. It combines - eigenface
- eigenUpper
- eigenTzone
- edge distribution
- These features are projected to a new
intra-person/extra-person similarity space and
are evaluated by a supporting vector machine
supervisor
21Matching/Classification
- Various matching schemes
- Neural Networks (NN)
- Deformable Models
- Hidden Markov Models (HMM)
- Support Vector Machines (SVM)14
- And a lot of hybrid schemes have been applied in
this field
22Decision Making
- How to make a reasonable decision out of the
following results - Traditional BT classification result
- Soft BT classification results
- gender
- ethnic
- eye/hair color
- age
- height
- weight
23Decision Making - Cont
- Approaches could be used
- Decision Tree
- Neural Network
- Bayesian approach
- Supporting vector machine
24System Diagram
25References
- 1 Anil K. Jain, Sarat Dass and Karthik
Nandakumar, Soft Biometric Traits for Personal
Recognition System. - 2 Anil K. Jain, Arun Ross and Salil Prabhakar,
An introduction to biometric Recognition, IEEE
Trans. on Circuits and Systems for Video
Technology, Special Issue on Image- and
Video-Based Biometrics, Vol. 14, No. 1, Jan.
2004. - 3 P. J. Philips, P. Grother, R. J. Micheals,
D. M. Blackburn, E. Tabassi, and J. M. Bone,
FRVT 2002 Overview and Summary, March 2003,
Available from http//www.frvt.org/FRVT2002/docum
ents.htm - 4 The Facial Recognition Technology (FERET)
Database, Available from http//www.itl.nist.gov
/iad/humanid/feret/feret_master.html - 5 Computer Vision Test Images, Available
from - http//www-2.cs.cmu.edu/cil/v-images.html
- 6 Jiebo Luo, Chang Wen Chen, Parker, K.J.,
Face location in wavelet-based video compression
for high perceptual quality videoconferencing,
Circuits and Systems for Video Technology, IEEE
Trans. on , Vol. 6 , No. 4 , Aug. 1996, pp 411
414. - 7 Chai, D., Ngan, K.N., Automatic Face
Location for Videophone Images, TENCON '96.
Proceedings. 1996 IEEE TENCON. Digital Signal
Processing Applications , Vol. 1 , 26-29 Nov.
1996, pp.137 - 140 vol.1
26Reference - Cont
- 8 Dugelay, J.-L. Junqua, J.-C. Kotropoulos,
C. Kuhn, R. Perronnin, F. Pitas, I. Recent
advances in biometric person authentication,
Acoustics, Speech, and Signal Processing, 2002.
Proceedings. (ICASSP '02). IEEE International
Conference on , Vol. 4, 13-17 May 2002, pp.
IV-4060 - IV-4063 vol.4 - 9 Zehang Sun Xiaojing Yuan Bebis, G. Louis,
S.J. Neural-network-based Gender Classification
using Genetic Search for Eigen-feature
Selection, Neural Networks, 2002. IJCNN '02.
Proceedings of the 2002 International Joint
Conference on , Vol. 3 , 12-17 May 2002, pp. 2433
2438 - 10 Gutta, S. Huang, J.R.J. Jonathon, P.
Wechsler, H. Mixture of experts for
classification of gender, ethnic origin, and pose
of human faces, Neural Networks, IEEE Trans. on
, Vol. 11 , Issue 4 , July 2000, pp. 948 960 - 11 Lanitis, A. Draganova, C. Christodoulou,
C. Comparing Different Classifiers for
Automatic Age Estimation, Systems, Man and
Cybernetics, Part B, IEEE Trans. on , Vol. 34 ,
Issue 1 , Feb. 2004, pp. 621 628 - 12 Hayashi, J. Yasumoto, M. Ito, H.
Koshimizu, H. Age and Gender Estimation based
on Wrinkle Texture and Color of Facial Images,
Pattern Recognition, 2002. Proceedings. 16th
International Conference on , Vol. 1 , 11-15 Aug.
2002, pp. 405 - 408 vol.1 - 13 Kwang In Kim Keechul Jung Hang Joon Kim
Face recognition using kernel principal
component analysis, Signal Processing Letters,
IEEE , Vol. 9 , Issue 2 , Feb. 2002, pp. 40 42
27Reference - Cont
- 14 G. D. Guo, S. Z. Li, and K. L. Chan, Face
recognition by Support Vector Machines, in Proc.
Int. Conf. Automatic Face and Gesture
Recognition, 2000, pp. 196-201. - 15 Do-Hyung Kim Jae-Yeon Lee Jung Soh
Yun-Koo Chung Real-time face verification using
multiple feature combination and a support vector
machine supervisor, Acoustics, Speech, and
Signal Processing, 2003. Proceedings. (ICASSP
'03). 2003 IEEE International Conference on ,
Vol. 2 , 6-10 April 2003, pp. II - 353-6 vol.2