research article palm-print pattern matching based on...

Research ArticlePalm-Print Pattern Matching Based on Features UsingRabin-Karp for Person Identification

S Kanchana1 and G Balakrishnan2

1Anna University of Technology Trichy Tamil Nadu India2Indra Ganesan College of Engineering Trichy Tamil Nadu India

Correspondence should be addressed to S Kanchana kanchanaarungmailcom

Received 20 April 2015 Revised 18 August 2015 Accepted 27 October 2015

Academic Editor Michele Nappi

Copyright copy 2015 S Kanchana and G Balakrishnan This is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

Palm-print based individual identification is regarded as an effectual method for identifying persons with high confidence Palm-print with larger inner surface of hand contains many features such as principle lines ridges minutiae points singular points andtextures Feature based pattern matching has faced the challenge that the spatial positional variations occur between the trainingand test samples To perform effective palm-print features matching Rabin-Karp Palm-Print Pattern Matching (RPPM) method isproposed in this paper With the objective of improving the accuracy of pattern matching double hashing is employed in RPPMmethod Multiple patterns of features are matched using the Aho-Corasick Multiple Feature matching procedure by locating theposition of the features with finite set of bit values as an input text improving the cumulative accuracy on hashing Finally a timeefficient bit parallel ordering presents an efficient variation on matching the palm-print features of test and training samples withminimal time Experiment is conducted on the factors such as pattern matching efficiency rate time taken on multiple palm-printfeature matching efficiency and cumulative accuracy on hashing

1 Introduction

Biometrics has been receiving a mushroom growth with theincrease in the civilian military and forensic applicationswhich involves with the significant identification of per-sons on the basis of biological or behavioral characteristicsNowadays a plethora of biometrics-based methodologiesis available and among them palm-print identification hasreceived greater attention Singular Value Decomposition(SVD) [1] based minutiae matching method involved inthe recognition of individuals through finger vein thoughresulted in improved identification of individuals with com-promised accuracy Two categories of minutiae called bifur-cation point (BP) and ending point (EP) were applied fromthe skeletonized binary finger vein images and extractedthree local descriptors local average intensity (LAI) localintensity deviation (LID) and local extensive binary pattern(LEBP) These three features were used for efficient minutiaematching Though the method was proven to be robust and

reliable accuracy of minutiae matching algorithm remainsunaddressed

Multimodal Sparse Representation (MSR) [2] methodreduced noise and occlusion through correlations and cou-pling information though it handled nonlinear variations butat the cost of accuracy which is included in our methodthrough Rabin-Karp based Palm-Print Pattern matchingalgorithm MSR presented a robust feature level fusion algo-rithm to handle different dimensions of different modalitiesthrough their sparse coefficientsThemethodwas also provedto be robust to occlusion and noise by introducing anoptimization framework that handled nonlinearity throughkernelizationThough qualitymeasure formultimodal fusionusing join sparse representation was handled in an efficientmanner the rate of accuracy remained unsolved

One of the major problems faced in person identificationbased on palm-print images is the individual identificationwith high confidence The significance of the problem liesin the effective feature matching using palm-print images

Hindawi Publishing Corporatione Scientific World JournalVolume 2015 Article ID 382697 8 pageshttpdxdoiorg1011552015382697

2 The Scientific World Journal

To address the problem related to feature (ie principlelines ridges minutiae points singular points and textures)matching higher accuracy in this work Rabin-Karp Palm-Print Pattern Matching (RPPM) method is presented Theproblems related to the existing method are that thoughreliability of palm-print matching was robust to noise andocclusion effective palm-print features matching was notaddressed In order to address an effective palm-print featuresmatching Aho-Corasick Multiple Feature matching proce-dure is applied in the proposed RPPMmethod

The contributions of our palm-print feature matchingalgorithm are summarized as follows First we learn differentangle of position using double hashing with finite set of bitvalues as an input text and avoid the hash collision to improvethe accuracy of pattern matching The value of palm-printpattern with different features is evaluated using the doublehashing which motivates us to exploit different angle ofposition and use it for palm-print feature matching Secondmultiple patterns of features between the test and trainingsamples are matched simultaneously using the Aho-CorasickMultiple Feature matching procedure It is different fromthe widely used multiple feature matching where the Aho-Corasick Multiple Feature matching procedure locates theposition of the features with finite set of bit values as an inputtext The finite set of bit values learnt from multiple featuresmakes our palm-print feature matching more effective butinvolves lengthy search process which is double the time ofthe actual single palm-print feature matching Third withan efficient bit parallel ordering a high level time efficientmethod is constructed on matching the palm-print featuresof test and training samples with minimal time

This paper is organized as follows Section 2 providesa review of the existing literature on palm-print biometricpattern matching Section 3 provides the details of the dou-ble hashing procedure and Rabin-Karp Palm-Print PatternMatching algorithm In Section 4 we present the experimen-tal setup with description of database In Section 5 resultsanalysis is discussed with parametric definitions FinallySection 6 includes the concluding remarks

2 Related Work

Latent palm-prints include certain amount of significantcritical evidential measure for applications related to forensicas it is probably estimated that about 30 percent of thelatents obtained through crime scenes are those of palm-prints Minutiae propagation algorithm introduced in [3]provided measures for reducing the time taken during palm-print matching Reconstruction algorithm designed in [4]provided an insight into different types of attacks usingfingerprint recognition system However the features con-sidered were limited which is solved by applying differentfeatures like principle lines ridges minutiae points andtextures in our method Robust key point detection usingScale Invariant Feature Transform (SIFT) [5] resulted inimprovement in retrieval performance but at the cost of timewhich has been addressed in ourmethodusingTimeEffectiveBit Parallel Ordering Another latent fingerprint matchingintroduced in [6] provided an insight into improving latent

matching accuracy using feedback paradigm However thefeatures considered were restricted which have been includedin our work

Segmentation and enhancement of palm-prints in [7]were designed with an objective of providing robustness andaccuracy using Total Variation decomposition model butat the cost of time Nonmatch Probability (NMP) [8] wasdesigned to improve the accuracy of matching being per-formed using fingerprint characteristics but did not considerdifferent features The above two said issues were addressedin our work by introducing different features and effectivetime model Multispectral Palm-Print Recognition (MPR)[9] was introduced to improve the personal identificationsystem using Minimum Distance Classifier (MDC) schemeand Weighted Majority Voting (WMV) algorithm Withthe objective of improving the personal identification usingleft and right palm-print images specialized algorithm wasdesigned using score-level fusion in [10] However accuracywas compromised which has been introduced in our workusing double hashing procedure

Several research works for palm-print recognition hasbeen performed with the aid of 2D palm images Thoughhigher amount of accuracy can be achieved using 2D palm-print recognition much 3D palm structural informationis lost 3D palm-print classification using global featureswas introduced in [11] with the objective of reducing thenoise However multiple features were not addressed whichis solved through our method that applies for multiplefeatures Morphological operation was applied in [12] withthe objective of improving the total success rate using featureslike principle lines wrinkles ridges singular points andminutiae points

A hybrid biometric cryptosystem introduced in [13]solved the problems and issues related to security against dif-ferent attacks using fuzzy commitment approach Howeverthe false positive rate increased with the increase in moresophisticated keys which is solved in our method using Aho-Corasick procedure Centered Discretion Hashing techniquedesigned in [14] provided insights into reducing the toleranceregion with the application of Principal Orientation Pattern(POP) and Orientation Pattern (OP)

To minimize the vulnerabilities related to palm-printbiometric template security model was introduced in [15]Descriptor Based Hough Transform (DHT) [16] includedmechanisms for improving the matching accuracy usingpalm-print matching algorithm However matching perfor-mance was compromised with differing features taken intoaccount Palm-print classification using 119896-nearest neighboris used in [17] addressing the issues with the objective ofreducing the noise However the method highly depends onthreshold value which is addressed through different featuresusing double hashing table

More specifically the recognition of palm-print is basedon either low resolution or high resolution of images Minu-tiae cylindrical code applied in [18] provided an efficientfeature extractionmethod for palm-print recognition Quan-tum algorithms introduced in [19] provided an exponentialspeed-up using Quantum Fourier Transform (QFT) withthe objective of increasing the matching accuracy In [20]

The Scientific World Journal 3

FeaturesPrinciple lines

RidgesMinutiae points

TexturesMatching test and

training sample patterns

Feature set

Input vector Test output

Figure 1 Pattern match based on features

Fractional Differential Algorithm (FDA) was applied toreduce the noise However accuracy was not achieved in theabove-stated algorithms To average the equal error rate in[21] Dempster-Shafer fusion theory was applied to uniquefeatures using mean curvature features

Based on the aforementioned methods and techniqueswe design an efficient pattern matching procedure for palm-print images using Rabin-Karp Palm-Print Pattern Matchingalgorithm

3 Rabin-Karp Palm-Print Pattern MatchingBased on Features

Biometric palm-print image pattern matching uses Rabin-Karp Palm-Print Pattern Matching (RPPM) method whichis constructed with the help of double hashing method withdifferent angle of position RPPM method is applied formultiple features matching using Aho-Corasick procedureand time efficient featurematching using bit parallel orderingThe concept of bit pattern matching with features suchas principle lines ridges minutiae points and textures isperformed simultaneously in RPPM method The patternmatching with the features is clearly shown in Figure 1

As shown in Figure 1 the pattern matching based onfeatures for palm-print pattern uses the aggregation of fea-tures using RPPM method Palm-print matching is carriedout by making use of the features with precise characteristicsto attain high accuracy rate The feature principle line istaken as a significant part on palm-print pattern matchingin our proposed method RPPM Minutiae points are basedon position path and direction of ridges Texture is used toseparate the feature into the region of interest and performsthe matching process

The RPPM forming several palm-print images ldquo119868rdquo usesthe double hashing procedure to retrieve the result withcollision-free hash values on different angle of test andtraining sample position The modifications in the positionand direction are also easily matched by Rabin-Karp throughdouble hashing Multiple feature matching is done effectivelyusing Aho-Corasick procedure The feature bit matching ofthe palm-print image uses the Time Effective Bit ParallelOrdering method to perform the matching nonlinearly toreduce the time complexity

Table 1 Double hashing table

Hash 1 (ℎ1) key Hash 2 (ℎ

2) key

Probe sequence of bits fordifferent palm-print

imagesTraining Image

11989611234 119896

2876 1101001001

1198961556 119896

23567 1011110101

11989615785 119896

2897 10101010001

The overall structural diagram of RPPMmethod is shownin Figure 2

As illustrated in Figure 2 test samples are given as input tothe RPPMmethod Initially the RPPM introduces the doublehashing technique to reduce the collision rate Collision issaid to occur because different angular position of minutiaepoints results in differing matching rate while working withthe test and training sample images So double hashingtechnique is applied to RPPM method with the objective ofreducing the collision rate

Subsequently feature based pattern matching is carriedout using the Aho-Corasick Multiple Feature The objectivebehind the application of Aho-CorasickMultiple Feature is topresentmultiple features while performing patternmatchingTo perform multiple pattern matching Aho-Corasick Multi-ple Feature is applied to the RPPM method With the searchbeing linear using Aho-Corasick Multiple Feature resultingin time complexity Time Effective Bit Parallel Orderingmethod is employed for the fast matching of the palm-printfeatures

31 Double Hashing Procedure The goal of the RPPMmethod is to identify the consistent and diverse features inpalm-print by performing the double hashing procedureDouble hashing procedure in RPPM resolves the hash col-lisions with different values in hash tableThe double hashingfor RPPMmethod is formularized as

Double Hashing

= 1198961[ℎ1(119894 119895)] + 119896

2[ℎ2(119894 119895)] mod 119879

(1)

In (1) two hash functions ℎ1and ℎ

2are used widely for

efficient identification of thematched featureswith ldquo119894rdquo and ldquo119895rdquorepresenting the pixels of palm-print images that are testedTo fetch the accurate results without collisions ldquo119896

1rdquo and ldquo119896

2rdquo

is the alternate keys (ie based on the positional change) usedwith the aid of ldquo119879rdquo that represents the hash table for palm-print feature matching

If the palm-print features are not matched with the singlehash value then the other hashing is carried outwith differenthash key

As a result ldquo119894rdquo and ldquo119895rdquo probe pixels of palm-print imageare analyzed through the double hash key values in Rabin-Karp Palm-Print Method Rabin-Karp Palm-Print uses thedouble hash function and is represented in Table 1

The double hashing table produces the sample form ofhash key used for the palm-print feature matching The hash


Double hashing

Test sample Training sampleRabin-Karp Palm-PrintPattern Matching

Hash collision avoidance

Multiple feature pattern matching

Aho-Corasickmultiple feature

Fast matching of multiple features

Time effective bit parallel ordering

Figure 2 Overall structural diagram of RPPMmethod

Principle lines TextureRidges Minutiae points

110 | 1001 | 00101 | 011

Figure 3 Probe sequences of bits

keys ldquoℎ1rdquo and ldquoℎ

2rdquo are placed for removing the collision reso-

lution on the palm-print biometric individual identificationThe sequence of bits for different series of palm-print imagefeatures is generated and placed in the table to perform thematching The training samples used in the table are probesequences to test the palm-print image features

32Multiple Feature PatternMatching Once the relevant fea-tures are identified using double hashing procedure the nextstep is to perform the efficient pattern matching for multiplefeatures Let us assume a palm-print image of length ldquo119899rdquowith the different feature pattern ldquo119891rdquo producing the best caseof result on matching multiple features simultaneously Theprobe sequence of bits for different image performs differentseparation to match the specific features The different setof bits helps to recognize (ie to match) to test and trainsamples and to produce more accurate biometric results forindividualsThe probe sequence of bits on the double hashingtable is shown in Figure 3

Figure 3 shows the probe sequences of bits The abovedouble hashing based probe sequence helps to fetch multiplefeature result accurately using RPPMmethod

321 Aho-Corasick Multiple Feature Aho-Corasick proce-dure automates the transition of pattern matching of featureswithout any backtracking process Aho-Corasick is con-structed with the double hashing table using RPPMmethodThis procedure finds the right function ldquo119891

119894rdquo to match the

palm-print feature pattern accurately Each row in the doublehash table identifies the hash key values ldquo1rdquo and ldquo2rdquo for thatspecific probe sequence of bits whereas the column pathindicates the sequence of key value ldquo1rdquo and ldquo2rdquo and probesequence bit respectively Followed by this the matchingprocess is formulized as follows

Feature Matching (FM) = 1198911(1198961 1198962)

1198912(1198961 1198962) 119891

119899(1198961 1198962)

(2)

The features uses the Aho-Corasick data structure formatching multiple features with the hash key values In thisway multiple feature matching operation is carried out withexplicit value in RPPMmethod

33 Fast Matching of Multiple Features In Aho-Corasickthe linear form of multiple features is carried out in RPPMmethod but the time complexity arises on matching thepalm-print features in a linear fashion To reduce the timecomplexity in our proposed work Time Effective Bit ParallelOrdering method is designed

331 Time Effective Bit Parallel Ordering RPPM methodeasily constructs a nonlinear automaton to improve thematching of the test and training sample image bit in a parallelfashion Bit parallel ordering technique in RPPM method


Figure 4 Sample palm-prints in CASIA database

is favorable to cut down the time taken on matching thefeatures Bit parallel ordering of the probe in RPPM methodproduces the suitable bisection for matching different com-bination of features in an effective manner and is formulizedas

BPO = min tim [FM119899] 997888rarr bit ordering (3)

Bit Parallel Ordering ldquoBPOrdquo of features for matchingnonlinearity mainly depends on the ordering format Theldquomin timrdquo denotes the minimum time taken on matchingmultiple palm-print features simultaneously The ordering ofthe bit palm-print image feature is formulized as

0 bit (119868)

=

120575timrarr119899119868119894119895 (119891match) = 1

120575timrarr119899+1119868119894119895 (119891 minus 1match) else otherwise

(4)

The time factor on ordering the bit before performingthe palm-print feature matching is provided in (4) Here thepalm-print image is represented in with ldquo119868rdquo and the pixels aredenoted as ldquo119894rdquo and ldquo119895rdquo in order to perform feature matchingThe features are matched with the ordered bits of varyingsize of ldquo119899rdquo images The length varied features which are notmatched are removed to perform the accuratematching withthe double hashing table

4 Experimental Evaluation

Rabin-Karp Palm-Print Pattern Matching (RPPM) methoduses MATLAB coding to perform palm-print matching Ini-tially the features of the palm-print with the positional anglesare mentioned for effective processing CASIA databaseconsists of 5502 palm-print images of both left and rightpalms with 8-bit gray level JPEG files confined from 312 usersas depicted in Figure 4

The pattern matching efficiency rate time taken onmultiple palm-print feature matching efficiency cumulativeaccuracy on hashing and false positive rate on matching thepatterns are the factors used for experimenting evaluationThe proposed method is compared against the existingmethods such as Singular Value Decomposition (SVD) [1]based minutiae matching method and Multimodal SparseRepresentation (MSR) [2] method

The pattern matching efficiency rate in RPPM is theamount of patterns efficiency matched using the doublehashing procedureThe pattern efficiency rate is measured interms of percentage and is the ratio of difference between thenumber of featuresmatched and features provided as input tothe total number of features provided Consider

PME =119899

sum

119894=1

(119891matched minus 119891119894)

119899 (5)

From (5) the pattern matching efficiency rate ldquoPMErdquo isperformed by the ratio of difference between the number offeatures matched ldquo119891matchedrdquo and features provided as inputldquo119891119894rdquo where ldquo119899rdquo denotes the total features provided as input

The higher pattern matching efficiency proves the efficacyof the method The time taken on multiple palm-printfeature matching is the amount of time taken to perform thefeature matching for multiple palm-print features using thebit ordering It is measured in terms of milliseconds (ms)Consider

Time for pattern matching

= Time (0 bit (119868)) (6)

The cumulative accuracy on hashing using RPPMmethod is the difference between the measured features tothe actual features Consider

CAH =119899

sum

119894=1

(Measured119891119894minus Actual

119891119894) (7)

The false positive rate on matching the patterns refers tothemeasure of good features falsely identified as bad featuresIt is measured in terms of percentage () Consider

FPR =(Good

119891minus Bad

119891)

Good119891

(8)

The false positive on matching the patterns is the ratio ofdifference between the good features and bad features to goodfeatures Low false positive rate on matching the patternsconfers the efficiency of the method

5 Results Analysis of RPPM

The Rabin-Karp Palm-Print Pattern Matching (RPPM)method is compared against the existing Singular Value


Table 2 Tabulation for pattern matching efficiency

Number of images Pattern matching efficiency ()RPPM SVD MSR

3 6536 5933 50326 7143 654 56399 7585 6982 608112 7235 6632 573115 7845 7242 634118 8133 753 672921 8575 7972 7071

Decomposition (SVD) [1] based minutiae matching methodand Multimodal Sparse Representation (MSR) [2] methodThe experimental results using MATLAB are analyzed anddisplayed with the aid of tables and figures given below

51 Scenario 1 PatternMatching Efficiency Table 2 shows thepattern matching efficiency over 21 different images providedas input usingMATLABThe changes in the patternmatchingefficiency are also being observed even in case of dissimilarimages However the pattern matching efficiency in anincreasing stage till 9 images was considered But with anincrease in the number of images to 12 the pattern matchingefficiency decreased and then increased to 15 images This isbecause of the different images gathered from both the maleand female As these images are not similar the changes inthe pattern matching efficiency are also being observed

Comparatively from Figure 5 the pattern matching effi-ciency is improved using the proposed method RPPM withthe application of double hashing procedure On workingwith the test and training sample images the analysis usesdifferent angular position of minutiae points and results inhigher pattern matching efficiency rate by 7ndash9 comparedto SVD [1] In addition using RPPM method based on thepositional changes of the features with the aid of alter keyand hash table the features are matched not only with thesingle hash value but with different hash key resulting inthe improvement of pattern matching efficiency by 17ndash23compared to MSR [2]

52 Scenario 2 Time for Pattern Matching The convergenceplot for 21 images is depicted in Figure 6 and Table 3 Wecould observe that the proposed RPPM method achievedminimum time for pattern matching when compared toother methods We also figure out that in Figure 6 theproposed RPPMmethod shows an increase in the beginningof the convergence graphs with the setting of images withupdated training and test database during the early iterationsHowever when the number of images was 15 the time forpattern matching reduced in a drastic manner because of theBit Parallel Ordering method

Figure 6 shows that the time for pattern matchingincreases with the increase in the number of images andshows that a drift decrease occurs when 15 images wereusedThe time taken onmultiple palm-print featurematching

0

20

40

60

80

100

3 6 9 12 15 18 21

Patte

rn m

atch

ing

effici

ency

rate

()

Number of images

RPPMSVDMSR

Figure 5 Impact of pattern matching efficiency

RPPMSVDMSR

0

20

10

40

30

60

50

80

70

3 6 9 12 15 18 21

Tim

e for

pat

tern

mat

chin

g (m

s)

Number of images

Figure 6 Impact of time for pattern matching

efficiency is reducedwith the application of Time Effective BitParallel Ordering method

The Time Effective Bit Parallel Ordering method inRPPM effectively constructs a nonlinear automaton in aparallel manner for the test and training sample images byproducing suitable bisection and therefore reducing the timetaken on multiple palm-print feature matching by 20ndash26compared to SVD [1]

Moreover the length varied features of test and trainingsample images that are notmatched are removed usingRPPMand as a result the time taken is reduced on multiple palm-print feature matching by 19ndash44 compared to MSR [2]


Table 3 Tabulation for time for pattern matching

Number of images Time for pattern matching (ms)RPPM SVD MSR

3 36 45 526 42 53 509 48 59 6612 55 66 7315 46 57 6418 52 63 7021 55 66 73

Table 4 Tabulation for cumulative accuracy on hashing

Number of users Cumulative accuracy on hashing ()RPPM SVD MSR

5 5583 498 407510 6145 5642 473715 5835 5232 452720 6588 5985 518025 6235 5632 492730 6845 6242 553735 7388 6785 598

53 Scenario 3 Cumulative Accuracy on Hashing TheRabin-Karp Palm-Print Pattern Matching (RPPM) method is com-pared with the two existing methods in terms of cumulativeaccuracy on hashing in this section and is depicted in Table 4with differing samples The number of users ranges from 5 to35 where the experiments were conducted using MATLABWe can notice that the proposed RPPM method had bettercumulative accuracy on hashing compared to the state-of-the-art works respectively

From Figure 7 we can notice that the RPPM methodconverge high accuracy on hashing than SVD [1] andMSR [2]which increases the performance measure The cumulativeaccuracy on hashing is improved with the application ofmultiple feature pattern matching This is effectively carriedout using probing sequence of bits for different imageswhere efficiency performs different separation to match thespecific features using double hashing table This results inthe increase of cumulative accuracy on hashing using RPPMmethod by 8ndash10 compared to SVD and 19ndash27 comparedto MSR respectively

54 Scenario 4 Impact of False Positive Rate Convergencecharacteristics of measure of false positive rate for 35 testimages with varying principle lines ridges minutiae pointsand textures are considered and compared with two othermethods and are shown in Table 5

The targeting results of false positive rate on matchingthe patterns using RPPM method are compared with twostate-of-the-artmethods [1 2] In Figure 8 visual comparisonis presented based on the initialization of features Ourmethod differs from the SVD [1] and MSR [2] We have

Table 5 Tabulation for false positive rate

Number of users False positive rate ()RPPM SVD MSR

5 0135 0146 015710 0149 0159 017015 0158 0169 018020 0165 0176 018725 0155 0166 017730 0160 0171 018235 0168 0179 0190

RPPMSVDMSR

0

20

10

40

30

60

50

80

70

5 10 15 20 25 30 35

Cum

ulat

ive a

ccur

acy

on h

ashi

ng (

)

Number of users

Figure 7 Impact of cumulative accuracy

incorporated competent procedure called the Aho-Corasickprocedure The Aho-Corasick procedure designs the finitestate machine in an accurate manner for performing easymatching functions for multiple features without any back-tracking process and therefore minimizes the false positiverate on pattern matching using RPPM method by 6ndash8compared to SVD Furthermore the linear form of multiplefeature matching operation further enhances the accuracyand therefore reduces the false positive rate on patternmatching by 13ndash16 compared to MSR

6 Conclusion

The conventional palm-print based person identificationusually designed for providing high quality pattern matchingusing differing features like principle lines ridges minutiaepoints singular points and textures with high confidencemay not give satisfactory result for accuracy during patternmatching To improve the accuracy of palm-print featuresmatching and reduce the time taken on multiple palm-print


RPPMSVDMSR

0

005

01

02

015

5 10 15 20 25 30 35

False

pos

itive

rate

()

Number of users

Figure 8 Impact of false positive rate

features matching efficiency on palm-print images Rabin-Karp Palm-Print PatternMatching (RPPM)method based onDouble Hashing procedure and enhancing multiple featurematching using Aho-Corasick Multiple Feature matchingprocedure has been implementedThe three-step model fea-ture identification using Double hashing procedure multiplefeature patternmatching using Aho-Corasick procedure andfast matching of multiple features using time effective BitParallel Ordering method introduced in RPPM resulted insignificant improvement over the state-of-the-art methodsin terms of pattern matching efficiency time on multiplepalm-print feature matching efficiency cumulative accuracyon hashing and false positive on matching the patterns

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

References

[1] F Liu G Yang Y Yin and S Wang ldquoSingular value decom-position based minutiae matching method for finger veinrecognitionrdquo Neurocomputing vol 145 pp 75ndash89 2014

[2] S Shekhar V M Patel N M Nasrabadi and R ChellappaldquoJoint sparse representation for robust multimodal biomet-rics recognitionrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 36 no 1 pp 113ndash126 2014

[3] E Liu A K Jain and J Tian ldquoA coarse to fine minutiae-based latent palmprint matchingrdquo IEEE Transactions on PatternAnalysis andMachine Intelligence vol 35 no 10 pp 2307ndash23222013

[4] J Feng and A K Jain ldquoFingerprint reconstruction fromminutiae to phaserdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 33 no 2 pp 209ndash223 2011

[5] U Park J Park and A K Jain ldquoRobust keypoint detectionusing higher-order scale space derivatives application to imageretrievalrdquo journal of IEEE Signal Processing Letters vol 21 no 8pp 962ndash965 2014

[6] S S Arora E Liu K Cao and A K Jain ldquoLatent finger-print matching performance gain via feedback from exemplarprintsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 36 no 12 pp 2452ndash2465 2014

[7] K Cao E Liu and A K Jain ldquoSegmentation and enhancementof latent fingerprints a coarse to fine ridge structure dictionaryrdquoIEEE Transactions on Pattern Analysis andMachine Intelligencevol 36 no 9 pp 1847ndash1859 2014

[8] ANagarHChoi andAK Jain ldquoEvidential value of automatedlatent fingerprint comparison an empirical approachrdquo IEEETransactions on Information Forensics and Security vol 7 no6 pp 1752ndash1765 2012

[9] S Minaee and A Abdolrashidi ldquoOn the power of joint wave-let-DCT features for multispectral palmprint recognitionrdquohttparxivorgabs14097818

[10] Y Xu L Fei andD Zhang ldquoCombining left and right palmprintimages for more accurate personal identificationrdquo IEEE Trans-actions on Image Processing vol 24 no 2 pp 549ndash559 2015

[11] P A Mane and A S Gaikwad ldquo3D palm print classificationusing global featuresrdquo International Journal of Advance Researchin Computer Science andManagement Studies vol 2 no 7 2014

[12] S D Raut and V T Humbe ldquoBiometric palm prints featurematching for person identificationrdquo International Journal ofModern Education and Computer Science vol 4 no 11 pp 61ndash69 2012

[13] A Nagar K Nandakumar and A K Jain ldquoA hybrid biometriccryptosystem for securing fingerprint minutiae templatesrdquoPattern Recognition Letters vol 31 no 8 pp 733ndash741 2010

[14] S Mundada P Wankhade S Kolte and S Konde ldquoPalm printidentification using centered discretization hashing techniquerdquoInternational Journal of Engineering Research amp Technology vol3 no 2 2014

[15] AK Jain andKNandakumarBiometric Authentication SystemSecurity and User Privacy IEEE Computer Society 2012

[16] A A Paulino J Feng and A K Jain ldquoLatent fingerprint match-ing using descriptor-based hough transformrdquo IEEE Transac-tions on Information Forensics and Security vol 8 no 1 pp 31ndash45 2013

[17] P A Mane and A S Gaikwad ldquoA novel approach to palmprintclassification using global featuresrdquo International Journal ofEmerging Technology and Advanced Engineering vol 4 no 102014

[18] P V Dudhanale and S R Ganorkar ldquoStudy of person identifi-cation using palmprint recognition system based on minutiaecylindrical coderdquo International Journal of Research in Engineer-ing and Technology vol 3 no 6 pp 536ndash539 2014

[19] H Li and Z Zhang ldquoResearch on palmprint identificationmethod based on quantum algorithmsrdquo The Scientific WorldJournal vol 2014 Article ID 670328 8 pages 2014

[20] C Chi and F Gao ldquoPalm print edge extraction using fractionaldifferential algorithmrdquo Journal of Applied Mathematics vol2014 Article ID 896938 7 pages 2014

[21] J Ni J Luo and W Liu ldquo3D palmprint recognition usingDempster-Shafer fusion theoryrdquo Journal of Sensors vol 2015Article ID 252086 7 pages 2015

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



To address the problem related to feature (ie principlelines ridges minutiae points singular points and textures)matching higher accuracy in this work Rabin-Karp Palm-Print Pattern Matching (RPPM) method is presented Theproblems related to the existing method are that thoughreliability of palm-print matching was robust to noise andocclusion effective palm-print features matching was notaddressed In order to address an effective palm-print featuresmatching Aho-Corasick Multiple Feature matching proce-dure is applied in the proposed RPPMmethod

The contributions of our palm-print feature matchingalgorithm are summarized as follows First we learn differentangle of position using double hashing with finite set of bitvalues as an input text and avoid the hash collision to improvethe accuracy of pattern matching The value of palm-printpattern with different features is evaluated using the doublehashing which motivates us to exploit different angle ofposition and use it for palm-print feature matching Secondmultiple patterns of features between the test and trainingsamples are matched simultaneously using the Aho-CorasickMultiple Feature matching procedure It is different fromthe widely used multiple feature matching where the Aho-Corasick Multiple Feature matching procedure locates theposition of the features with finite set of bit values as an inputtext The finite set of bit values learnt from multiple featuresmakes our palm-print feature matching more effective butinvolves lengthy search process which is double the time ofthe actual single palm-print feature matching Third withan efficient bit parallel ordering a high level time efficientmethod is constructed on matching the palm-print featuresof test and training samples with minimal time

This paper is organized as follows Section 2 providesa review of the existing literature on palm-print biometricpattern matching Section 3 provides the details of the dou-ble hashing procedure and Rabin-Karp Palm-Print PatternMatching algorithm In Section 4 we present the experimen-tal setup with description of database In Section 5 resultsanalysis is discussed with parametric definitions FinallySection 6 includes the concluding remarks

2 Related Work

Latent palm-prints include certain amount of significantcritical evidential measure for applications related to forensicas it is probably estimated that about 30 percent of thelatents obtained through crime scenes are those of palm-prints Minutiae propagation algorithm introduced in [3]provided measures for reducing the time taken during palm-print matching Reconstruction algorithm designed in [4]provided an insight into different types of attacks usingfingerprint recognition system However the features con-sidered were limited which is solved by applying differentfeatures like principle lines ridges minutiae points andtextures in our method Robust key point detection usingScale Invariant Feature Transform (SIFT) [5] resulted inimprovement in retrieval performance but at the cost of timewhich has been addressed in ourmethodusingTimeEffectiveBit Parallel Ordering Another latent fingerprint matchingintroduced in [6] provided an insight into improving latent

matching accuracy using feedback paradigm However thefeatures considered were restricted which have been includedin our work

Segmentation and enhancement of palm-prints in [7]were designed with an objective of providing robustness andaccuracy using Total Variation decomposition model butat the cost of time Nonmatch Probability (NMP) [8] wasdesigned to improve the accuracy of matching being per-formed using fingerprint characteristics but did not considerdifferent features The above two said issues were addressedin our work by introducing different features and effectivetime model Multispectral Palm-Print Recognition (MPR)[9] was introduced to improve the personal identificationsystem using Minimum Distance Classifier (MDC) schemeand Weighted Majority Voting (WMV) algorithm Withthe objective of improving the personal identification usingleft and right palm-print images specialized algorithm wasdesigned using score-level fusion in [10] However accuracywas compromised which has been introduced in our workusing double hashing procedure

Several research works for palm-print recognition hasbeen performed with the aid of 2D palm images Thoughhigher amount of accuracy can be achieved using 2D palm-print recognition much 3D palm structural informationis lost 3D palm-print classification using global featureswas introduced in [11] with the objective of reducing thenoise However multiple features were not addressed whichis solved through our method that applies for multiplefeatures Morphological operation was applied in [12] withthe objective of improving the total success rate using featureslike principle lines wrinkles ridges singular points andminutiae points

A hybrid biometric cryptosystem introduced in [13]solved the problems and issues related to security against dif-ferent attacks using fuzzy commitment approach Howeverthe false positive rate increased with the increase in moresophisticated keys which is solved in our method using Aho-Corasick procedure Centered Discretion Hashing techniquedesigned in [14] provided insights into reducing the toleranceregion with the application of Principal Orientation Pattern(POP) and Orientation Pattern (OP)

To minimize the vulnerabilities related to palm-printbiometric template security model was introduced in [15]Descriptor Based Hough Transform (DHT) [16] includedmechanisms for improving the matching accuracy usingpalm-print matching algorithm However matching perfor-mance was compromised with differing features taken intoaccount Palm-print classification using 119896-nearest neighboris used in [17] addressing the issues with the objective ofreducing the noise However the method highly depends onthreshold value which is addressed through different featuresusing double hashing table

More specifically the recognition of palm-print is basedon either low resolution or high resolution of images Minu-tiae cylindrical code applied in [18] provided an efficientfeature extractionmethod for palm-print recognition Quan-tum algorithms introduced in [19] provided an exponentialspeed-up using Quantum Fourier Transform (QFT) withthe objective of increasing the matching accuracy In [20]






Feature set











2) key



11989611234 119896

2876 1101001001

1198961556 119896

23567 1011110101

11989615785 119896

2897 10101010001





Double Hashing

= 1198961[ℎ1(119894 119895)] + 119896

2[ℎ2(119894 119895)] mod 119879

(1)





2rdquo






Double hashing









110 | 1001 | 00101 | 011











1198912(1198961 1198962) 119891

119899(1198961 1198962)

(2)









0 bit (119868)

=

120575timrarr119899119868119894119895 (119891match) = 1


(4)






PME =119899

sum

119894=1

(119891matched minus 119891119894)

119899 (5)




= Time (0 bit (119868)) (6)


CAH =119899

sum

119894=1


119891119894) (7)


FPR =(Good

119891minus Bad

119891)

Good119891

(8)







3 6536 5933 50326 7143 654 56399 7585 6982 608112 7235 6632 573115 7845 7242 634118 8133 753 672921 8575 7972 7071






0

20

40

60

80

100

3 6 9 12 15 18 21

Patte

rn m

atch

ing

effici

ency

rate

()

Number of images

RPPMSVDMSR


RPPMSVDMSR

0

20

10

40

30

60

50

80

70

3 6 9 12 15 18 21

Tim

e for

pat

tern

mat

chin

g (m

s)

Number of images








3 36 45 526 42 53 509 48 59 6612 55 66 7315 46 57 6418 52 63 7021 55 66 73



5 5583 498 407510 6145 5642 473715 5835 5232 452720 6588 5985 518025 6235 5632 492730 6845 6242 553735 7388 6785 598







5 0135 0146 015710 0149 0159 017015 0158 0169 018020 0165 0176 018725 0155 0166 017730 0160 0171 018235 0168 0179 0190

RPPMSVDMSR

0

20

10

40

30

60

50

80

70

5 10 15 20 25 30 35

Cum

ulat

ive a

ccur

acy

on h

ashi

ng (

)

Number of users



6 Conclusion



RPPMSVDMSR

0

005

01

02

015

5 10 15 20 25 30 35

False

pos

itive

rate

()

Number of users





References





























Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








Feature set











2) key



11989611234 119896

2876 1101001001

1198961556 119896

23567 1011110101

11989615785 119896

2897 10101010001





Double Hashing

= 1198961[ℎ1(119894 119895)] + 119896

2[ℎ2(119894 119895)] mod 119879

(1)





2rdquo






Double hashing









110 | 1001 | 00101 | 011











1198912(1198961 1198962) 119891

119899(1198961 1198962)

(2)









0 bit (119868)

=

120575timrarr119899119868119894119895 (119891match) = 1


(4)






PME =119899

sum

119894=1

(119891matched minus 119891119894)

119899 (5)




= Time (0 bit (119868)) (6)


CAH =119899

sum

119894=1


119891119894) (7)


FPR =(Good

119891minus Bad

119891)

Good119891

(8)







3 6536 5933 50326 7143 654 56399 7585 6982 608112 7235 6632 573115 7845 7242 634118 8133 753 672921 8575 7972 7071






0

20

40

60

80

100

3 6 9 12 15 18 21

Patte

rn m

atch

ing

effici

ency

rate

()

Number of images

RPPMSVDMSR


RPPMSVDMSR

0

20

10

40

30

60

50

80

70

3 6 9 12 15 18 21

Tim

e for

pat

tern

mat

chin

g (m

s)

Number of images








3 36 45 526 42 53 509 48 59 6612 55 66 7315 46 57 6418 52 63 7021 55 66 73



5 5583 498 407510 6145 5642 473715 5835 5232 452720 6588 5985 518025 6235 5632 492730 6845 6242 553735 7388 6785 598







5 0135 0146 015710 0149 0159 017015 0158 0169 018020 0165 0176 018725 0155 0166 017730 0160 0171 018235 0168 0179 0190

RPPMSVDMSR

0

20

10

40

30

60

50

80

70

5 10 15 20 25 30 35

Cum

ulat

ive a

ccur

acy

on h

ashi

ng (

)

Number of users



6 Conclusion



RPPMSVDMSR

0

005

01

02

015

5 10 15 20 25 30 35

False

pos

itive

rate

()

Number of users





References





























Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Double hashing









110 | 1001 | 00101 | 011











1198912(1198961 1198962) 119891

119899(1198961 1198962)

(2)









0 bit (119868)

=

120575timrarr119899119868119894119895 (119891match) = 1


(4)






PME =119899

sum

119894=1

(119891matched minus 119891119894)

119899 (5)




= Time (0 bit (119868)) (6)


CAH =119899

sum

119894=1


119891119894) (7)


FPR =(Good

119891minus Bad

119891)

Good119891

(8)







3 6536 5933 50326 7143 654 56399 7585 6982 608112 7235 6632 573115 7845 7242 634118 8133 753 672921 8575 7972 7071






0

20

40

60

80

100

3 6 9 12 15 18 21

Patte

rn m

atch

ing

effici

ency

rate

()

Number of images

RPPMSVDMSR


RPPMSVDMSR

0

20

10

40

30

60

50

80

70

3 6 9 12 15 18 21

Tim

e for

pat

tern

mat

chin

g (m

s)

Number of images








3 36 45 526 42 53 509 48 59 6612 55 66 7315 46 57 6418 52 63 7021 55 66 73



5 5583 498 407510 6145 5642 473715 5835 5232 452720 6588 5985 518025 6235 5632 492730 6845 6242 553735 7388 6785 598







5 0135 0146 015710 0149 0159 017015 0158 0169 018020 0165 0176 018725 0155 0166 017730 0160 0171 018235 0168 0179 0190

RPPMSVDMSR

0

20

10

40

30

60

50

80

70

5 10 15 20 25 30 35

Cum

ulat

ive a

ccur

acy

on h

ashi

ng (

)

Number of users



6 Conclusion



RPPMSVDMSR

0

005

01

02

015

5 10 15 20 25 30 35

False

pos

itive

rate

()

Number of users





References





























Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








0 bit (119868)

=

120575timrarr119899119868119894119895 (119891match) = 1


(4)






PME =119899

sum

119894=1

(119891matched minus 119891119894)

119899 (5)




= Time (0 bit (119868)) (6)


CAH =119899

sum

119894=1


119891119894) (7)


FPR =(Good

119891minus Bad

119891)

Good119891

(8)







3 6536 5933 50326 7143 654 56399 7585 6982 608112 7235 6632 573115 7845 7242 634118 8133 753 672921 8575 7972 7071






0

20

40

60

80

100

3 6 9 12 15 18 21

Patte

rn m

atch

ing

effici

ency

rate

()

Number of images

RPPMSVDMSR


RPPMSVDMSR

0

20

10

40

30

60

50

80

70

3 6 9 12 15 18 21

Tim

e for

pat

tern

mat

chin

g (m

s)

Number of images








3 36 45 526 42 53 509 48 59 6612 55 66 7315 46 57 6418 52 63 7021 55 66 73



5 5583 498 407510 6145 5642 473715 5835 5232 452720 6588 5985 518025 6235 5632 492730 6845 6242 553735 7388 6785 598







5 0135 0146 015710 0149 0159 017015 0158 0169 018020 0165 0176 018725 0155 0166 017730 0160 0171 018235 0168 0179 0190

RPPMSVDMSR

0

20

10

40

30

60

50

80

70

5 10 15 20 25 30 35

Cum

ulat

ive a

ccur

acy

on h

ashi

ng (

)

Number of users



6 Conclusion



RPPMSVDMSR

0

005

01

02

015

5 10 15 20 25 30 35

False

pos

itive

rate

()

Number of users





References





























Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






3 6536 5933 50326 7143 654 56399 7585 6982 608112 7235 6632 573115 7845 7242 634118 8133 753 672921 8575 7972 7071






0

20

40

60

80

100

3 6 9 12 15 18 21

Patte

rn m

atch

ing

effici

ency

rate

()

Number of images

RPPMSVDMSR


RPPMSVDMSR

0

20

10

40

30

60

50

80

70

3 6 9 12 15 18 21

Tim

e for

pat

tern

mat

chin

g (m

s)

Number of images








3 36 45 526 42 53 509 48 59 6612 55 66 7315 46 57 6418 52 63 7021 55 66 73



5 5583 498 407510 6145 5642 473715 5835 5232 452720 6588 5985 518025 6235 5632 492730 6845 6242 553735 7388 6785 598







5 0135 0146 015710 0149 0159 017015 0158 0169 018020 0165 0176 018725 0155 0166 017730 0160 0171 018235 0168 0179 0190

RPPMSVDMSR

0

20

10

40

30

60

50

80

70

5 10 15 20 25 30 35

Cum

ulat

ive a

ccur

acy

on h

ashi

ng (

)

Number of users



6 Conclusion



RPPMSVDMSR

0

005

01

02

015

5 10 15 20 25 30 35

False

pos

itive

rate

()

Number of users





References





























Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






3 36 45 526 42 53 509 48 59 6612 55 66 7315 46 57 6418 52 63 7021 55 66 73



5 5583 498 407510 6145 5642 473715 5835 5232 452720 6588 5985 518025 6235 5632 492730 6845 6242 553735 7388 6785 598







5 0135 0146 015710 0149 0159 017015 0158 0169 018020 0165 0176 018725 0155 0166 017730 0160 0171 018235 0168 0179 0190

RPPMSVDMSR

0

20

10

40

30

60

50

80

70

5 10 15 20 25 30 35

Cum

ulat

ive a

ccur

acy

on h

ashi

ng (

)

Number of users



6 Conclusion



RPPMSVDMSR

0

005

01

02

015

5 10 15 20 25 30 35

False

pos

itive

rate

()

Number of users





References





























Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




RPPMSVDMSR

0

005

01

02

015

5 10 15 20 25 30 35

False

pos

itive

rate

()

Number of users





References





























Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



research article palm-print pattern matching based on...

Documents