a high performance fingerprint matching system for large databases based on gpu

Download A High Performance Fingerprint Matching System for Large Databases Based on GPU

If you can't read please download the document

Upload: alpesh-kurhade

Post on 18-Jul-2015

84 views

Category:

Engineering


4 download

TRANSCRIPT

Slide 1

1Presented ByAlpesh KurhadeUnder The Guidance Of Mrs. M. A. Shah Seminar OnA High Performance Fingerprint Matching Systemfor Large Databases Based on GPU

Presented ByAlpesh D. KurhadeM.Tech I YearUnder the Guidance ofMrs. M. A. Shah1Introduction

Related Work

Proposed Work

System Flow and Methodology

Conclusion

Future Work

References

2ContentsDEFINITION OF BIOMETRICS

Biometrics refers to the automatic identification of a person based on his or her physiological or behavioral characteristics 4BIOMETRIC TERMSRecognitionIdentificationVerification

Fingerprint recognitionA live acquisition of a persons fingerprint.

Dots (very small ridges),

Bifurcation,

It can be characterized through some particular called minutiae.

Biometrics IdentificationBiometricData CollectionTransmissionSignal Processing, Feature Extraction, RepresentationQuality Sufficient?Generate Template Template MatchDecisionConfidence?YesNoDatabase

new biometric sample is requested. NoYes

Minutiae Matching.

ContinueMatching process is the main bottleneck of identification system.Local minutiae matching algorithm: - define neighborhood.

Global minutiae matching algorithm: - use the information of all the minutiae at once.Continue.2) Minutia Cylinder Code algorithm (MCC)9

This algorithm uses only the position and orientation of minutiaeThe similarity is defined by:

The cylinder Ca and Cb are matchable if the directional difference between two minutiae is not greater than a certain value.What is GPU?It is used in embedded systems, mobile phones, personal computers, workstations, and game consoles.Highly parallel structure makes them more effective.

A Single Instruction Multiple Data (SIMD) architecture is used in GPU devices to introduce parallelism

Some GPU are:#1 GeForce GTX 580#2 GeForce GTX 480#3 Quadro 6000#4 GeForce GTX 570#5 GeForce GTX 470#6 Radeon HD 687011

Structure Used By GPUThreadBlocksGridsWraps

A GPU-Based AlgorithmThe adaptation of the different data structures.

Calculation of the algorithm on the GPU.

Specific enhancements for identification systems.Data structures are one of the key issues.

Use coalesced memory access

3. Each fingerprint(minutiae)Float4 data type4. Fingerprint Database constructed using two linear arraysData StructuresComputationCylinder Generation:- The number of cells per cylinder (Ns Ns Nd ) is one of the MCC algorithm parameters.

Fig:graphic scheme of the computational structure152. Fingerprint Matching.Continue.Comparison between input fingerprint to a set of fingerprints stored in a database.

Processes has been adapted to the parallel GPU architecture.

Performance Enhancements for Identification SystemsA fingerprint identification systems goal is not to perform one to one fingerprint matches but to find the matching fingerprint in a database to match an input fingerprint.The reduction of GPU idle periods.The packaging of several matching processes into one.

Reducing the GPU idling periodsScalabilityMulti-GPU

Fingerprint Database SizeEXPERIMENTAL RESULTSExperiments

Hardware

Empirical ResultsTwo different types of GPUs have been used in the experiments:

Tesla GPU, an NVIDIA Tesla M2090 with 512 CUDA cores, Fermi architecture and 6GB of memory.

2) GTX GPU, an NVIDIA GeForce GTX 680 with 1536 CUDA cores, Kepler Architecture and 2GB of memory.Hardware

Empirical Results

Fingerprints per second with the DB4 database.ContinueAn efficient GPU based Fingerprint method using MCC algorithm.

Obtained speed-up ratios up to 100.8 with respect to a single-thread CPU implementation.

System has no scaling issues when the fingerprint database sizeIncreases.

d) It is able to perform an identification in a reasonable amount of time for large databases, processing up to 55,700 fingerprints per second with a single GPU.

23ConclusionTo study other aspects of the fingerprint identification.

The reduction of database fingerprint candidates.

Use of several fingerprints from the same person in the identification process24Future ScopeReferences25Pablo David Gutirrez, Miguel Lastra, Francisco Herrera, and Jos Manuel Bentez , A High Performance Fingerprint Matching System for Large Databases Based on GPU, IEEE Trans. On Information Forensics And Security , Vol. 9, No.1, Jan 2014.

D. Maltoni, D. Maio, A. Jain, and S. Prabhakar, Handbook of Fingerprint Recognition. New York, NY, USA: Springer-Verlag, 2009.

R. Cappelli, M. Ferrara, and D. Maltoni, Minutia cylinder-code: A new representation and matching technique for fingerprint recognition, IEEE Trans. Pattern Anal. Mach. Intell., vol. 32, no. 12, pp. 21282141, Dec. 2010.

Lin Hong, Student Member, IEEE, Yifei Wan, and Anil Jain, Fellow, IEEE., Fingerprint Image Enhancement :Algorithm and Performance Evaluation, IEEE Trans On pattern analysis & machine intellegence, vol. 20, no. 8, aug.1998.

.

26 Thank You