Upload
nguyentuyen
View
218
Download
0
Embed Size (px)
Citation preview
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Modris Greitans, Mihails Pudzs, Rihards FuksisInstitute of Electronics and Computer ScienceDzerbenes 14, Riga, Latviae‐mail: [email protected]
The 12th ACM Workshop on Multimedia and SecuritySeptember 9‐10, Rome, Italy
Research is supported by:
ESF project Nr. 1DP/1.1.1.2.0/09/APIA/VIAA/020, co‐financed by EU
Latvian State research program in innovative materials and technologies
Palm Vein Biometrics Based on Palm Vein Biometrics Based on Infrared Imaging and Complex Infrared Imaging and Complex
Matched FilteringMatched Filtering
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Motivation
•Identity fraud•Linking physical person to a digital identity
•Biometrics
ProblemProblem Solution Solution
2/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Why Palm Veins?
•Invisible in daylight •Hard to falsify•Unique structure•Allow distinguish twins•Easy to use
3/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Outline
Image acquisitionImage acquisition ProcessingProcessing RecognitionRecognition ResultsResults
4/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Outline
Image acquisitionImage acquisition ProcessingProcessing RecognitionRecognition ResultsResults
4/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Outline
Image acquisitionImage acquisition ProcessingProcessing RecognitionRecognition ResultsResults
?
4/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Outline
Image acquisitionImage acquisition ProcessingProcessing RecognitionRecognition ResultsResults
4/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Imaging MethodsReflection Transmission
AdvantagesAdvantages DisadvantagesDisadvantagesAdvantagesAdvantages DisadvantagesDisadvantages
Low power consumptionCompact design
Low power consumptionCompact design
Some wrinkles visible
Some wrinkles visible
High power consumptionBulky design
High power consumptionBulky design
Slightly better palm vein visibility
Slightly better palm vein visibility
5/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Acquired Images
Visible light Reflectionmethod
Transmissionmethod
6/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Vessel Analysis
Cross section
Vessel cross section can be approximated with Gaussian function
7/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Matched Filtering
MatchedFiltering
⎩⎨⎧
+=−−=
φφσφ
sincos')2/'exp(),( 22
yxxxyxg x
8/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security 8/20
Matched filtering
MatchedFiltering
⎩⎨⎧
+=−−=
φφσφ
sincos')2/'exp(),( 22
yxxxyxg x
Matched filtering ‐ SLOW
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Complex Matched Filtering
=),( yxCMF ⎥⎦
⎤⎢⎣
⎡⋅∑ ),,(2
ll
j yxGe l ϕϕ
ϕ2je⋅ ∑
For further information: M.Greitans, M.Pudzs, R.Fuksis. „Object Analysis in Images Using Complex 2d Matched Filters”, Proceedings of the IEEE Region 8 Conference EUROCON 2009. Saint–Petersburg, Russia, May, 2009., pp. 1392‐1397.
9/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
CMF Result
CMF
Advantages:Filtering with one complex maskAdditional information about the vessel’s orientationDrawback: Halo effect
Advantages:Filtering with one complex maskAdditional information about the vessel’s orientationDrawback: Halo effect
10/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Building Vector Sets
After CMF we can construct the most significant vector set
240x320 pixels240x320 pixels 128 bytes128 bytes
11/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Building Vector Sets
Iterative execution ofthe following steps:
•Find max response•Save the vector•Exclude neighborvectors from further processing
•Continue
12/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Building Vector Sets
Iterative execution ofthe following steps:
•Find max response•Save the vector•Exclude neighborvectors from further processing
•Continue
12/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Building Vector Sets
Iterative execution ofthe following steps:
•Find max response•Save the vector•Exclude neighborvectors from further processing
•Continue
12/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Building Vector Sets
Iterative execution ofthe following steps:
•Find max response•Save the vector•Exclude neighborvectors from further processing
•Continue
12/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Comparison of Vectors
3 Impact factors:magnitudes angles distance
Calculation of similarity:1.Pair of longer vectors have greater influence on similarity2.Closely oriented vectors have greater impact3.Closer the vectors – more considerable contribution
13/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Comparison of Vector Sets
Similarity index of two vector sets:
),(),(),(),(
AAsBBsABsABS⋅
=
Normalization
Drawbacks:1.Index is influenced by image contrast2.Neighborhood vessel impact on similarity
Drawbacks:1.Index is influenced by image contrast2.Neighborhood vessel impact on similarity
Advantages:1.Value is normalized between 0 and 1Advantages:1.Value is normalized between 0 and 1
Algorithm is not rotation and scale invariant – we have used the palm fixing stand for accurate image acquisition without shifts
14/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Construction of the Database
•Database of 400 images from 50 persons
•Result of comparison for 2 cases
FA FR
15/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Comparison with Database
6464 6464
??
Vector set
6464 6464 6464 6464 6464
6464 6464 6464 6464 6464 6464 6464
6464 6464 6464 6464 6464 6464 6464
6464 6464 6464 6464 6464 6464
6464 6464 6464 6464 6464 6464 6464
6464 6464 6464 6464 6464 6464 6464
6464 6464 6464 6464 6464 6464 6464
6464DatabaseDatabase
Because of the time critical recognition stage we evaluate performance with several number of vectors ‐ 64 / 51 / 26
16/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Results of database evaluation
Each image is compared with each other image in database
Each black square represents the 8 images of each person
Black dots represent the FAWhite dots represent the FR
17/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Experimental results
P2PM ICPM
EER 2.679% 0.557%
Our method
EER 0.17%
These methods* use binary image comparison and the database of 500 persons
Our method uses vector set comparison, but database of 50 persons
* H. Chen, G. Lu, and R. Wang. A new palm vein matching method based on icp algorithm. In ICIS'09: Proceedings of the 2nd International Conference on Interaction Sciences, pages 1207–1211, New York, NY, USA, 2009. ACM.
18/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security
Conclusions
19/20
Rihards Fuksis The 12th ACM Workshow on Multimedia and Security 1/xRihards Fuksis The 12th ACM Workshop on Multimedia and Security 20/20