18-03-2010, 08:03 AM
[attachment=2712]
Fingerprint Recognition :Future Directions
Presented BY:
Salil Prabhakar
Digital Persona Inc.
Fingerprint Applications
Commercial
Computer Network Logon,Electronic Data Security,E-Commerce,InternetAccess,ATM, Credit Card,Physical Access Control,Cellular PhonesPersonal Digital Assistant,Medical Records,Distance Leaning, etc.
Government
National ID card,Correctional Facilities,Driverâ„¢s License,Social Security,Welfare Disbursement,Border Control,Passport Control, etc
Forensic
Corpse IdentificationCriminal Investigation,Terrorist Identification,Parenthood determination,Missing Children, etc.
Fingerprint Application Functionality
Positive Identification
Is this person truly know to the system
Commercial applications (network logon)
Desirable: low cost and user-friendly
Large Scale Identification
Is this person in the database
Government and Forensic applications (prevent double dipping; multiple passports)
Desirable: high throughput with little human intervention
Surveillance and Screening
Is this a wanted person
Airport watch list
Fingerprints are not suitable
Reasons for Accuracy Challenges
Information Limitation Due to individuality, poor presentation and inconsistent acquisition.
Design and choice of representation (features) and quality of feature extraction algorithms (especially for poor quality fingerprints)
Invariance Limitation
Incorrect modeling of invariant relationships among features
Fingerprint Individuality Estimation
Accuracy; Information Limitation
Assumptions for theoretical individuality estimation consider only minutiae (ending and bifurcation) features minutiae locations and directions are independent minutiae locations are uniformly distributed correspondence of a minutiae pair is an independent event quality is not explicitly taken into account ridge frequency is assumes to be constant across population and spatially uniform in the same finger analysis of matching of different impressions of the same finger binds the parameters of the probability of matching prints from different fingers an alignment between two fingerprints has been established.
Probability of a False Correspondence
Accuracy; Information Limitation; Fingerprint Individuality Estimation
m = no. of minutiae in template
n = no. of minutiae in input
= no. of corresponding minutiae based on location (x,y) alone
q = no. of corresponding minutiae based on location and direction ()
A = area of overlap between input and template
C = area of tolerance region = r02/A
Probability that one of one input minutiae matches any of the m template minutiae:
Probability that two of two input minutiae matches any of the m template minutiae:
Information Limitation: Conclusion
Accuracy; Information Limitation
There is an incredible amount of information content in fingerprints
A minutiae-based fingerprint identification system can distinguish between identical twins
The performance of state-of-the-art automatic fingerprint matchers do not even come close to the theoretical performance
Performance of fingerprint matcher is depended on the fingerprint class and thus may depend upon target population
Fingerprint classification may not be very effective in genetically related population
Fingerprint identification accuracy may suffer in certain demographics
Conventional Representations
Accuracy;
Minutiae-based
Sequential design based on the following modules: Segmentation, local ridge orientation estimation (singularity and more detection), local ridge frequency estimation, fingerprint enhancement, minutiae detection, and minutiae filtering and post-processing.
Ridge Feature-based
Size and shape of fingerprint, number, type, and position of singularities (cores and deltas), spatial relationship and geometrical attributes of the ridge lines, shape features, global and local texture information, sweat pores, fractal features.
Representations: Future Directions
Accuracy;
Improvement of current representations through robust and reliable domain-specific image processing techniques such as:
Model-based orientation field estimation
Robust image enhancement and masking New richer representations Fusion of various representations
Matching: Future Directions
Accuracy; Invariance Limitation
Alignment remains a difficult problem “ develop alignment techniques that remain robust under the presence of false features
Understand and model fingerprint deformation
Fusion of various matchers (based on the same or different representations)
Multiple Biometrics; Fusion
A decision (and lower) level fusion of multiple biometrics can improve performance
In identification systems, fusion can also improve speed
Independence among modalities is key
Even combination of correlated modalities can be no worse than the best performing modality alone
Best combination scheme would be application dependent
Performance Evaluation
Evaluation types: technology, scenario, operational
Dependent on composition of the population (occupation, age,demographics, race), the environment, the system operational mode, etc
Ideally, characterize the application-independent performance in laboratory and predict technology, scenario, and operational performances
Standardization and independent testing
Parametric and non-parametric estimation of confidence intervals and database size
Parametric and non-parametric and statistical modeling of inter-class and intra-class variations;
Usability, Security, Privacy
Biometrics are not secrets and not revocable
Encryption, secure system design, and liveness detection solve this problem
Unintended functional scope; unintended application scope; covert acquisition
Legislation; self-regulation; independent regulatory organizations
Biometric Cryptosystems: fingerprint fuzzy vault
Alignment
Similarity metric in encrypted domain
Variable and unordered representation
Performance loss; ROC remains the bottleneck