Image-Based Biometric Person Authentication
Post on 15-Jan-2016
27 Views
Preview:
DESCRIPTION
Transcript
Professor Heikki Kälviäinen
1 Machine Vision and Pattern Recognition Laboratory
Image-Based Biometric Person Authentication
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory (MVPR)
Department of Information Technology
Faculty of Technology Management
Lappeenranta University of Technology (LUT)
Heikki.Kalviainen@lut.fi
http://www.lut.fi/~kalviai
http://www.it.lut.fi/ip/research/mvpr/
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
2
Content
• Machine vision and pattern recognition in LUT. • Biometric person authentication.• Face detection.• Why is detection/localization difficult.• Existing approaches.• Proposed algorithm.• Results and evaluation.• New solutions. • Conclusions.
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
3
Machine Vision and Pattern Recognition Laboratory (MVPR)
Leader: Prof. Heikki Kälviäinen. 2nd largest computer vision research group in Finland. Center of Excellence in Research in LUT.24 members:
• 3 Professors + 3 Post docs + 2 Visiting doctors + 11 PhD students + undergraduate students + industry coordinator.
Co-operation with 14 international universities and research institutes.Results: 18 Ph.D. degrees (and 3 externally produced), over 400 scientific
publications, 40 research projects, and spin-off companies. Objectives: 2 PhDs/year. Annual external project funding 700.000 EUR, basic funding 300.000 EUR,
total 1.0 million EUR.http://www.it.lut.fi/ip/research/mvpr/
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
4
MVPR Laboratory: Research Profile
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
5
Machine Vision System
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
6
Hand geometry
Iris
FingerprintsFace recognition
BiometricPerson
Authentication
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
7
Biometric: any measurement of a person’s physiological traits or behavior
Physiological:• Face• Fingerprint• Iris• Retinal scan• Ear shape• Hand geometry• Infrared (face, body parts)• Odor
Behavioral:• Speech• Handwriting• Signature• Lip movements• Keystroke dynamics• Gait
Genetic:• Tissue sample
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
8
FACEDETECT Image-Based Biometric Person Authentication
http://www.it.lut.fi/project/facedetect/
Docent, Dr. Joni Kamäräinen, Docent, Dr.Ville Kyrki, Mr. Pekka Paalanen, Mr. Jarmo Ilonen,
Prof. Heikki KälviäinenMachine Vision and Pattern Recognition Research Group
Lappeenranta University of TechnologyFINLAND
Dr. Miroslav Hamouz, Prof. Josef Kittler,
Prof. Jiri Matas
Centre for Vision, Speech, and Signal Processing (CVSSP)
University of Surrey
UNITED KINGDOM
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
9
Why is Face Detection Difficult?
• Object-class recognition (an object to be recognized is not a single entity rather a a group of similar objects).
• Faces exhibit significant variability in shape, colour, and texture, and may appear in arbitrary poses:
– Appearance variations over the whole population.
– Capture effects.
– Background.
• Illumination.
• Video versus still image.
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
10
State of the Art
• Image-based methods: – Scanning window. – Face modeled as manifolds in some high dimensional space.
Moghaddam, Pentland – probabilistic PCA
Sung and Poggio, Rowley et al.- neural networks
Osuna et al. – SVM
Viola and Jones – Adaboost on
Haar features
Jesorsky et al – Haussdorf
distance on edge images
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
11
State of the Art (cont.)•Feature-based methods:–Face modelled as a viable configuration of local features.
–Needs higher resolution than image-based methods.
–False alarms.
Vogelhuber, Schmid, Gaussian derivatives + angles and length ratios Weber er al., interest operator + statistical model on positionsCristinacce and Cootes, Adaboost + shape model
•Warping methods: Variability decomposed into a shape model and the model of local appearance or texture which is iteratively deformed to fit.
Cootes et al., Active Shape and Appearance modelsLades et al., Wiskott et al. Dynamic link architectures
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
12
Face verification (authentication) Validating a claimed identity based on the image of a face: are you Mr./Ms. X?
Face recognition (identification)Identifying a person based on an image of his/her face: who are you?
Face detection/localizationLocation of human faces in images at different positions, scales, orientations, and lighting conditions.
Introduction
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
13
Proposed Algorithm
• Avoiding a scanning window.• Using feature detectors.• Shape-free texture model for the final decision.
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
14
Feature Detector: 2-D Gabor Filter
cossin'
sincos'
),( '2''22
2
22
2
2
yxy
yxx
eef
yx fxjy
fx
f
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
15
Gabor Features
• Maximal joint localization in the spatial and frequency domain.• Smooth and noise tolerant.• Parameters for invariance manipulation:Frequency Envelope sharpness Orientation
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
16
Constructing Response Matrix
Filter response r(x,y; f,) can be calculated for variousfrequencies f and orientations to construct a response matrix.
columns represent orientationsrows represent frequenciesimage rotation appears as acircular shift of the columns
image scaling appears as ashift of the rows (highfrequencies may vanish)
A SCALE AND ROTATION INVARIANTTREATMENT OF THE RESPONSE MATRIXCAN BE ESTABLISHED, AND THUS, WECAN CONCENTRATE ONLY HOW TOCLASSIFY THEM IN THE STANDARD POSE
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
17
2-D Gabor Features
discrete frequency [u]
dis
cre
te fr
eq
ue
ncy
[v]
-1/2 -1/4 0 1/4 1/2
-1/2
-1/4
0
1/4
1/2
What do they ”see”?
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
18
Evidence Extraction
Requirements
• Scale invariant extraction.• Rotation invariant extraction.• Provides sufficiently small amount of correct candidate points. (n best points from each class; needs confidence measure).
Preferred
• Estimation of evidence scale and orientation.• Fast extraction (scalability).
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
19
Classifier Construction
eye
eye
nostrilnostril
eyeeye
Gaussian mixturemodel densities(EM estimation)
• Stability property guarantees approximately the Gaussian form of classes in the feature space.
• One class may still consist of several sub-clusters (open eye, closed eye, etc.).
Bayesianclassificationof features
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
20
Affine Learned Correspondences
Aligned images of objects andmanually selected features Variability and correspondences
1 2
3 4
5 6
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
21
Affine Hypothesis Search
2
2
3 11 12
4
5
2
2
1
1. Evidence extraction.
2. Affine search and match to correspon- dence model.
Instanceapproved
False alarms occur and hypothesisverification is needed
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
22
Face Space
• Normalization of space where shape variations and capture effects are removed from patterns.
• Based on three points on the face -> affine registration.
• Optimal with regard to the photometric variance over a big set of faces.
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
23
Features & Feature Detectors
• Features = salient parts of face.
• Small localization variance and frequent occurrence over population.
• Illumination, scale, rotation, and translation invariance.
• Automatic analysis using the face space desirable.
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
24
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
25
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
26
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
27
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
28
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
29
Confidence Regions
• Exhaustive search over triplets O(n)n3.
• Not all triplets have to verified, regions supporting highly likely transformations can be learned.
• Speed-up up to 1 000 times.
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
30
Performance Measure
rl
rleye CC
ddd
),max(
• Strict measure using the location of eye centres, not only an upright bounding box.
• deye<=0.05 in order to succeed in verification.
• deye<=0.25 corresponds to the definition of successfuldetection in the majority of state-of-the-art algorithms.C = ground truth eye center coordinates
d = distances between the detected
and ground truth ones
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
31
deye = 0.05
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
32
Recognition System
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
33
BANCA Database
Large realistic face and voice database collected (BANCA database):• 4 languages, each language 6540 images of 52 people.• Three scenarios simulating controlled access, office environment and
outdoor scenes.• Publicly available including a rigorous evaluation protocol.
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
34
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
35
XM2VTS Database
LABEL Rate (%)
1 56.1
2 84.2
3 70.9
4 50.9
5 84.9
6 64.2
7 70.4
8 75.5
9 54.2
10 45.8
1 triplet detected(%)
88.3
Both eye centres detected (%)
74.5
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
36
BioID Database
LABEL Rate (%)
1 55.6
2 67.6
3 51.2
4 39.1
5 61.1
6 54.8
7 29.5
8 34.5
9 40.0
10 48.7
1 triplet detected(%)
73.4
Both eye centres detected (%)
48.6
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
37
BANCA Database
LABEL Rate (%)
1 41.4
2 60.3
3 44.5
4 44.0
5 67.4
6 34.3
7 54.8
8 63.6
9 49.6
10 61.8
1 triplet detected(%)
81.4
Both eye centres detected (%)
44.0
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
38
3-dimensional Face Recognition
• 3-D images.• 3-D algorithms.• Accurate!
• Images? • Reference
databases?• Speed?
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
39
FACEDETECT - Publications
Hamouz, M., Kittler, J., Kamarainen, J.-K., Paalanen, P., Kälviäinen, H., Matas, J., Feature-Based Affine-Invariant Localization of Faces, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 27, No. 9, 2005, pp. 1490-1495. (Impact factor: 3.810)
Kamarainen, Joni-Kristian, Ville Kyrki, and Heikki Kälviäinen. Invariance properties of Gabor filter based features - Overview and applications. IEEE Transactions on Image Processing, Vol. 15, No. 5, 2006, pp. 1088-1099. (Impact factor: 2.428)
Kyrki, Ville, Joni-Kristian Kamarainen, and Heikki Kälviäinen. Simple Gabor feature space for invariant object recognition. Pattern Recognition Letters, Vol. 25. No. 3. 2004, pp. 311-318. (Impact factor: 1.138)
Paalanen, P., Kamarainen, J.-K., Ilonen, J., Kälviäinen, H., Feature Representation and Discrimination Based on Gaussian Mixture Model Probability Densities - Practices and Algorithms, Pattern Recognition, Vol. 39, No. 7, 2006. pp.1346-1358. (Impact factor: 2.153)
Professor Heikki Kälviäinen
Machine Vision and Pattern Recognition Laboratory
40
Conclusions and Future Work
• Algorithm successfully tested on a large face authentication data set.• Combination of features brings a significant performance boost.• Gabor jets proved as a suitable local representation of a signal.• Adequate resolution necessary for feature detectors to succeed.• 3-D face recognition much more accurate than 2-D recognition.
• Methods for non-frontal poses (more 3-D face research needed).• Speed: real-time solutions (3-D image acquisition and analysis).• Applications:
– Security applications: biometric passports, access, cash dispensers, etc.
– Surveillance applications.
top related