state of art analysis
TRANSCRIPT
Politecnico di MilanoDipartimento di Elettronica, Informazione e Bioingegneria (DEIB)
Biomed Meeting
Sara [email protected]
Thursday, April 7, 2016
Giulia [email protected]
State of Art Analysis
2
Easy to use
Non invasive
From any vital subject
Accessible to various sites
Low cost
Less fragile
PPG Biometric
3
2003First PPG recognition methods
2007PPG derivatives
2011Automated feature
extraction
2013Continuos
authentication Methods
2014New features-
ranking algorithm
2015Samsunginterest
2015APG
4
Methods:
PPG sensor attached to the fingertip
Statistical features extraction based
Machine learning based
Statistical consistency
Discriminability
Goals:
Feature extraction methods
Acquisition:
5
[1] Y. Y. Gu, Y. Zhang, and Y . T. Zhang, “A Novel Biometric Approach in HumanVerification by Photoplethysmographic Signals”, 2003
[2] Y. Y. Gu, Y. Zhang, and Y . T. Zhang, “Photopletismographic Authentication trough Fuzzy Logic”, 2003
First approach
17 subjects
94% success
Derivatives
3 subjects
[3] J. Yao, X. Sun, and Y. Wan, “A Pilot Study on Using Derivatives of Photoplethysmographic Signals as a Biometric Identifier”, 2007
2003
2007
6
2003
2007
[4] P. Spachos, J. Gao and D. Hatzinakos, “Feasability study of photopletysmographic signals for biometric authentication”, 2011
Automated way2011
2 datasets 14/15 subjects
Automatic features extraction
Linear Discriminant Extraction (LDA)Feature extraction tool + Supervised learning methods
7
[4] P. Spachos, J. Gao and D. Hatzinakos, “Feasability study of photopletysmographic signals for biometric authentication”, 2011
OpenSignal PPG Datasets BioSec PPG Datasets
Clustering
PPG signals must be obtained in a controlled environment and with accurate sensors
8
2003
2007
[5] A. Bonissi, R. D. Labati, L. Perico, R. Sassi, F. Scotti, L. Sparagino, “A preliminary study on continuous athentication methods for phoplethysmographic biometrics”, 2013
Continuous authentication methods2011
44 subjects
2013
2 min
Segmentation 20s, 30s, 40s
14 subjects
15 min
Segmention40s
9
[5] A. Bonissi, R. D. Labati, L. Perico, R. Sassi, F. Scotti, L. Sparagino, “A preliminary study on continuous athentication methods for phoplethysmographic biometrics”, 2013
2 min 15 min
Best results with 40s
Features time variability
Continuous enrollment needed
10
2003
2007
[6] A. R. Kavsaoglu, K. Polat, M. R. Bozkurt, “A novel feature algorthm for biometric recognition with ”, 2014
A novel feature-ranking algorithm2011
2013
2014
30 subjects
1st configuration
2nd configuration
3rd configuration
11
2003
2007
[6] A. R. Kavsaoglu, K. Polat, M. R. Bozkurt, “A novel feature algorthm for biometric recognition with ”, 2014
2011
2013
2014
Feature-ranking (FR) algorithm
1 2 3 … … … 39 40
k- nearest neighborsValidation
A novel feature-ranking algorithm
30 subjects
40 features
12
[6] A. R. Kavsaoglu, K. Polat, M. R. Bozkurt, “A novel feature algorthm for biometric recognition with ”, 2014
k n° features Success with FR
(%)
Success without FR
(%)
1st configuration 1 25 90,44 89,33
2nd configuration 1 20 94,44 90,22
3rd configuration 3 15 87,22 84,22
Best results with Feature-Ranking
Poor results considering time evolution
13
2003
2007
[7] A. Lee and Y. Kim, “Photoplethysmography as a form of biometric recognition”, 2015
2011
2013
2014
10 subjects
2015
Feel-forward neural networks
PPG biosignals contain uniquely identifiable information
PPG sensitivity to physical conditions
22 features
http://www.samsung.com
14
[8] K. A. Sidek, N. I. Zainal, S. N. A. M. Azam and N. A. L. Jaafar, “The development of human biometric identification using acceleration plethysmogram”, 2015
APG
2003
2007
2011
2013
2014
2015
Acceleration plethysmogram
15
Year Authors Innovation Methods Subjects Results in recognition
2003 Y. Y. Gu et al.
PPG for biometrics
Statistical 17 Success 94%
2007 J. Yao et al. PPG derivatives
Statistical 3
2011 P. Spachos et al.
Automated feature extraction
Machine learning
14/15 EER 25%
2013 A. Bonissi et al.
Continuous authentication
Statistical 44 2 min EER 5.29%15 min EER 13,47%
2014 A. R. Kavsaoglu et al.
Features-ranking algorithm
Machine learning
30 Success 94,44%Success 87,22%
2015 A. Lee et al. Samsung’s interest
Machine learning
10 Success 95,88%
2015 K. A. Sidek et al.
APG Machine learning
10 Success 98%
16
2003
2007
2011
2013
2014
2015
2016
Wearable device
ECG-based algorithms
Enhance accuracy
Strong biometric recognition system
[email protected]@mail.polimi.it
Emails
https://www.facebook.com/bioreds.project/
Politecnico di Milano, NECST lab, DEIB, building 20, via Ponzio, 34/5, 20133, Milano
https://twitter.com/BioREDs_necst