Making Palm Print Matching Mobile

(IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 Making Palm P rint Matching Mobile Fang L i School of computer engin eering Nan yang T echno logic al U nive rsit y Sing apore asfli @nt u.edu. sg Maylor K.H. Leung School of computer engin eering Nan yang T echno logic al U nive rsit y Sing apore asmkl eun g@ntu. edu. sg Chen g Shao Chia n School of computer engin eering Nan yang T echno logic al U nive rsit y S ingapor e Y060043@ ntu .edu.sg Abstract — With the grow ing import ance of per sonal ide ntifi catio n an d authe ntic at ion in to day’ s hig hly a dvan ced w orld where most business an d personal tas ks are being replac ed by electroni c means, th e need for a technol ogy that is a ble to uniq uely ident ify an in divi dual and has hig h f raud r esi stanc e s ee the rise of biometric te chnologi es. Mak i ng b iometric - based solut ion mob ile is a pr omising trend . A ne w RST i nvari ant squar e - bas ed pal m print ROI extractio n method was successfully imple ment ed a nd i ntegr ated i nt o the c urr ent a ppli catio n sui te . A new set of palm pr int i mage data bas e capt ure d using e mbedde d cameras in mobile phone w as created to test its robust ne ss. C om pari ng to th ose ex traction meth ods that are base d on bound ary trac king of the over all hand sh ape th at has li mitati on of bei ng una ble to p roce ss pa lm pri nt i mages t hat has one o r more fingers cl osed, the sy stem can n ow effectively handle the seg mentat ion of pal m pri nt i mages w it h varyi ng fi nger posit ioni ng. The high fle xi bili ty makes pal m print mat ching mobile possible. Keyword s - Palm p rint , segmen tation , mobility ; I. I NTRODUCTION Persona l identificat ion and authe ntication have beco me a common ta sk in today’s hig hly adv ance d w orl d w here more and mor e day - to - day persona l and busi ness activities ha ve been co mputer ized [1 - 3] . Traditional i dentificati on a nd authenticati on system s rely on either a token ite m (e.g. a securit y pass card) or s ome kno wledge only the user w ould know (e. g. passwords). Such systems are usually expensi ve in terms of time and res ources t o maintai n and exp and its usa ge. The most crit ical flaw of these sy stems is that sin ce they d o not use any inh erent charac teristi cs or attri butes of t he in divid ual user, the y are unable to differentiate between a n authorized personne l and an impostor who have fra udulently come to p ossess t he token or knowledge (such as stole n credit card or lost pass word). As such, these problems have led to system develope rs and resear chers to explore into alt ernative solutions, and thus the intensified r esearch arises on biom etri c identifi cation and authent ication systems [4 - 6] . Followin g this initial f oray into biom etric research, several form s of bi om etric sy st em s bas ed on dif ferent p hysiological or behavi oral ch aract erist ics h ave b een d evelo ped. The fi rst comm ercial s ys tem, I dentim at w as d evelo ped in the 197 0s [ 7]. The sy stem was b ased on th e m easu rement of th e shap e of th e hand and the lengths of the fingers as the basis for pers on al identificati on. After that, various for ms of biometri c systems such as finge rprint - based sy stem s and ir is, re tina , face, pal m print, vo ice, handwriting a nd DNA technolo gies joined in over the y ears. Among the le ading biometric technologies, finge r print - based sy stem is the most promin ent and widely used biom etric technol ogy, encompas sing a market share of 58 % in 2007 (A combine p ercentage o f fingerpr int and AFIS/ Livescan technol ogies) [1]. The small size of the fingerpr int - bas ed device , ease of us e and h igh ac cura cy has made it l argel y popular; ho wever, as w ith most biometr ic solutions, it a lso ha s certain d raw backs . It is c ommonly f ound in most people that a layer of oil se cretion o r pers piration w hich em its from microscop ic pores residi ng on the tiny ridges of the finger s will cover the surface of the fingerpri nt areas. As the resolutio n required for the fingerprint images ar e relatively high at appro ximately 500 dpi [7], this layer of secre tion will render the fingerprint image capturing device use less or less effect ive in m ost cases . There are also cases w h ereby fi ngerp rints w ear away due to w ork or fraudule ntly s carred, all these will low er the effect iveness of fingerpri nt based systems. In this project, we explore a relatively new biometr ic techn ology that em ploys palm print as the physiologic al charact erist ic t hat is us ed t o d iff erentiate b etw een each uni que indivi dual. Pa lm prin ts are r ich in featu res such as p rincipa l lines, wrinkles, ridge , datum point s and minutiae po ints, all o f which co uld be e xtr acte d at re lati ve lo w reso lut ion . Palm pr int s also h ave a m uch la rger su rface area as com pared t o fingerpr ints, which indicate s that more features could be extract ed fr om i t, ad ding high er lev el of accu racy to it . These adva ntages place palm print - based technology as a promising biometric i dentificati on system . Palm print recognition is an effective biom etric technol ogy that is gaini ng w ides pread accept ance an d int erest fr om researchers all over the w orld. As w ith most other bi ometric technol ogies, the process of palm pri nt identification inc ludes various stages fr om data acqui siti on, data p re - processing, featur e extract ion to matchin g proces s. The m ain aim of this research is to im prove the (Region O f Inte rest ) ROI extr act ion process t o inc rease the sy stem robust ness. By implementing and integra ting a new square - based p alm print ROI method into t he previo us applicatio n suite, the system is now able t o overcom e the lim iting problem of failure to process palm print im ages with c losed finge rs, at the s ame t ime, to be (Rotation, Scaling and T ranslation) RST invariant , thus i ncrease the flexibility of the sy stem and in turn 1 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 open up the possibility of bringing the palm print technology mobile. A new set of palm print im age databas e c aptured using embedde d cam eras i n m obile phone w as creat ed t o find most robust ROI extr action technique s. II. O VERV I EW OF P A LM P R IN T M ATCHI NG S YSTE M Palm print recognitio n is an effective biometric te chnology that is gaining wi despread acceptance and in terest fr o m researchers all over th e world. As w ith most other biom etric technologies, t he process of pal m print identifica tion includ es various sta ges from data a cquisition, d ata pre - pr oce ssing, feature extraction to matching process. T he s ystem o ver vie w is sho wn in F igure 1. Fig ur e 1 . Palm prin t matchin g system over view . III. I MAGE A CQU I SIT ION In image acqui sition stage, an image of the user ’s palm is captur ed by the sy stem . Palm im age can be acquir ed by a f ew methods . One m ethod i s to a pply a unifor m layer of ink on t he palm and plac e the palm on a paper; the paper i s then scanned into P C to obtain a digital image . This method i s regard ed as an “o ff - line” metho d. Another metho d for palm image acquisitio n is t o use digital devices t o ph otogra ph th e palm and imm ediatel y obt ain a digital image s tored in the system. Such digital device can be a digital cam era [8 ] , which is shown in Figure 2, or sc anner [ 9 ], which is sho wn in Figure 3. This method is regarded a s an “on - line ” method. Figure 2 . Acquisition of a ty pical image s ample using digital camera [8] . Figure 3 . Acquisition of a ty pica l image sample using scanner [9]. Inked pa lm print is not a good choice for mobile matching, so we only discuss online pal m print in this research . IV. P RE - PROCESSING :R EG ION O F I NTER E ST (R OI) E XTRACTION After im age acquisiti on, the r aw input is passed to the verifi cation stages to per form various image proc essing operations. Normally , the raw image c onsists of palm, fin gers, wrist, and a substantial amount of bac kground area. For the verification process, only the inner a rea of the palm is of interes t. The system needs to trim away those unwanted portio ns of the raw im age to red uce the amount of computa tion required in the subsequent stages. Another problem with the raw input image is that the location and orientation of the palm is not fixed. Palm prints may show certain degree of dist orti on as the imag e may be c apture d at diffe rent ti mes and rotat ed at differ ent an gl es . F urther mor e, it co uld al so be affect ed by varying co nditions in terms of temperatur e, humidity and 2 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 lighting co ndition . A s s uch, even i f two images a re from the sam e palm, we could end up with a co nclusion that the y are fro m different so urces. Palm print prepro cessing, th e segm entati on pro cess, involvi ng the cor rect ion of su ch dis to rti on an d pl aci ng all t he palm prints in the database u nde r th e sam e coo rdinate sy stem and orie ntation such tha t the proper e xpected are a of each palm print can be extrac ted f or u se i n a ccurate feat ure extr action and matching, gr eatly impro ves the ef fici ency and co rrectn ess of the identif ication sy stem. The ou tpu t of p al m pr int s egmentation is a sub image , known as the R e gion of I nteres t (ROI) or cent ral part sub - image of the palm print , w hich is cut out from the original input i mage. This sub image repre sents the inner area of the palm, w here m ost of th e palm pr int f eatures are with in this area. Tho se method s can be further classified into two diffe rent class es: squ are - ba sed ROI extr action and inscri bed cir cle - based ROI extraction [8 ]. As the ci rcle - based appro ach co nsumes a significa ntly higher amount of c omputa tion res ources, based on this experimen tal outset, w e will only focus on square - bas ed ROI extraction approach throughout this researc h. The basi c idea of square - bas e d ROI e xtr actio n tec hnique as demonstrated in Figu re 4 is to deter mine key gaps - betw een - fin ger s point on a p alm print , thereaf t er t wo s elec ted key p o int s are lined up to for m the y - axis, s ubsequently , a se cond l ine, which is the x - axis is dra wn perpe ndicular to the y - axis thr ough the midd le point to form the origi n. Final ly , a squar e w ith a fixe d s ize, ROI , is extract ed under this c oordinate system. All the pixel s w ithin this ROI are retained for further proce ssing wher eas th e area outs ide the wind ow is ignor ed and discard ed. The es sential r ule in this e xtractio n process i s that the portion of th e image extracte d should be available in all palm pri nts from t he databas e an d the re are s uff icient palm prin t f eatur es for e xtraction and c omparison. Moreove r, the extraction should be RST invari ant and gesture differe nce tolerant. Figure 4. Basic idea of square - based segmentation technique . All RO I e xtrac tio n tec hniques principal ly rely on the determinati on of the key gaps - betw een - fingers to draw up the coordina te system through the use of boundary tracking algorithm. Thus this im poses a lim itati on in tha t for a palm print image to be prop erly segmented, the fingers in the image need to be sufficie ntly separated in order for the bound ary tracing t o work, and thus t he ke y points are det ermin ed accurat ely . Such require ment led to the developme nt of i mage acquisiti on devices , which are shown in Figure 5, 6, and 7, t hat utilizes pegs [10 - 13] to restrict the move ment and positioning of the ha nd during the acquisition proce ss in order to improve the image quali ty, to solve RST problem, and to ensur e th at the fingers a re proper ly separated . Figure 5. P alm image acquisiti on de vic e wi t h p eg s [ 10 ]. Figure 6. Pal m image acquisiti on d evic e with pegs and fixed ROI location [11]. Po lyU -O nline - Pa lm print - II is a benchmark palm print da tabas e for research purpos e [ 5] . In the data base, all th e palm imag es are captu red by a sp ecially desi gned d evice [1 3] as shown in Figur e 7 . The dista nce f rom palm to camera , pe gs between fing ers , and the bo ard between middle and ring fingers are fi xed. A lot of res e archers use this b enchmark d atabase and the dev ice const rain t ROI e xtra ction p rocess based on this databa se is shown in Figure 8 [12]. 3 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 Figure 7 . P atent Palm image acquis ition device designed by D. Zhang et al [13]. Figure 8. ROI extraction process based on database Poly U -O nline - Palm pri nt - II [12]. Howev er, such dev ices are n orm ally fix ed to a site, and too bulky to move around for usage . This lack of mobility conseq uently results in the restricti ve applications of palm print authenticati on system [14 - 16] . In t his project , so me built in cam era s in m obi le phon es are used for image acquisition. The ad vantage of using tho se built in camera s as input device s is that the user doe s not need to purchas e any device becaus e of the popu larity of m obile phone s. N ow ada ys , al mos t e ve ryon e h as a m ob ile ph one wi th built in cam era . U si ng thes e cam era s , the pal m print verification system can b e eas ily integr ated in to any security systems without requiring extra im age ac quisition dev ices. V. I MPROVE D S Q UARE - BAS ED ROI E XTRACTIO N M ETHOD T he a im of th is resea rch is to im plement an im proved ROI extracti on technique t hat is robust enough to overco me this relianc e of a s tanda rd im age ac quisiti on d evic e, an d thu s abl e to make use of the ubiq uitous d igital cam eras an d em bedd ed camera s in m obile phone s to perform the image capturing proce ss, which in turn, will widen the scope of applications for palm pri nt - bas ed system s. The i mpr ove d s quar e - bas ed ROI ex tr action technique consists of the foll owin g ste ps: Ste p 1: Gray im age to bina ry im age. Step 2: Co ntour o f hand gene rat ion The above two st ep are standard on es in all palm print preprocessing system, so details are omi tted here. Step3: Stra ight Lines Extracti on As mentioned in sec tion 4 , the locations of three feature points need to be detected in order to set up a coordinate system for pa lm print align ment. These key points lie on the bot tom of va lle ys bet ween f inger s. B y obser ving t he line pattern of th e boundary image, th e bottom of valley i s a short curve joi ning t he ed ges o f adj ace nt fi ngers. The key p oints ar e best represented as th e mid - po ints o f those shor t cur ves. To locate the mid - point, one metho d is to first find the li ne ( Lm ) tha t di vides the i nter - finge r spa ce i nto ha lves , t he intersecting point between Lm and the bott om c urves of the valle ys is one of the desir ed ke y point s. Us uall y, the edge s o f two adjacent fingers form a V - shape. An angle can be estab lis hed b y extend ing t he V - shap e ed ges unt il th ey intersect. The line Lm can be found b y calculat ing the bisector of suc h angle . T he met hod is illust rate d in Fi gure 9. 4 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 Fig ur e 9. Metho d of L o cating K ey Point . Generally, it is not easy to locate the V - shape edges directly. The problem can be solved by fin ding the paral lel edges of each f inger first, th en for every tw o adjacent fingers, select an appropriate edge from each finger, and use the two selected edges to form a V - shape pair. This approach is feasible since detecti ng parallel line pairs are m uch easi er than det ecti ng V - shape line pairs directly. Based on the method above, the task of locating the three key points in the boun dary image c an be div ided in to four steps: Step 3. 1: Straight line s extraction: Fi nd the straig ht lines i n the edge image, select the long o nes whic h have the p ote ntial to be edges of fing ers. Step 3. 2: Par allel lines grouping: Gr oup the extracte d long lines into parallel pairs, each pair represents two edges of a finger . Step 3. 3: V - Shape lines g rouping: Reorder the parallel pair s and gr oup t he line s into V - shape pairs, each pair represents the edges of two adjacent fingers. Step 3. 4: Key point s loca tion: Form an angle for e ach V - shape pair and calculate the bisector of each an gle. F ind the intersecting points between the boundaries of inter - fin ger valleys and the calculated bisectors. T he intersecting p oints represent the desired k ey points. Straigh t lines extraction The edge pixels of th e binary image are traversed an d the contour representation of the edge image is generated. Contour is a com pact way to represent the shape of an image. The edge pixels are form ed into separate g roups, where each group represents a connected curve. The Dynamic T wo - Str ip (DYN2S) algo rithm [ 17 ] is em ploy ed to perf orm the curv e fitting oper ation. If a curve has onl y small variation (i.e. it is closed to a straight line), then this curve will be reduced to a single line se gment which app roximates to the original c urve. After pro cessi ng, t he nu mber of line se gment s in t he ima ge i s greatly decreased, wh ich reduces the difficulty of ident ifyin g the finger edges . In or der to get lo ngest po ssib le li nes whi ch have the hi gh chance to represent the l ocation of the fingers , broken line conne ctio n algo ri thm descr ibe in [1 8 ] is adopted here to off set the pos sibl e broken lin es issue. The deta ils of th e boun dary of the palm are not n ecessary to an alyze, so tiny straight lines are excluded. Parall el Line s Groupi ng Finding the par allel pairs i n this method mea ns f indi ng the two lines which are the edge of the f inger. To find the parallel lines, first ly o ne li ne mus t be t aken a nd the n che ck e very ot he r line whether it is the para llel partner of the first line or not. After that, the next line must be taken as the first line, a nd check the other li ne to find it s parallel partner. It m ust be done until all of t he li nes have b een che cke d whet her t hey have parallel p artner or not. T he extracted parallel pairs extracted fro m the pa lm ima ge in F igur e 9 are s hown in F ig ure 10. Fig ur e 10 . The p arallel pai rs . V- shaped Line s Groupi ng V- shap ed line s group in g mea ns dete ctin g the two line in between two fingers. Fir stly, the parallel p airs obtained from the pre vious algori thm must be sorte d from the l eftmost to the rightmost. Since the parallel pairs obtained from the previous algor ithm a re s tore d usi ng the 2 - D arra y, it is easy to sor t the parallel lines and get the V - shaped line pairs. To get the V - shaped pair s, it is basically shi fting all the lines b y one in to t he right as belo w: - Sort the para llel line pairs, so that t he line pairs are stored in left to rig ht order. - For each parallel pair Pi in the sorted array, f orm a V - shape pair with the righ t edge of P i and the left edge of P i+1 (i = 0. .I - 2, where I is th e total number of paralle l pairs). The result of this p hase can b e see n in t he F i gure 11 . T he same color line s identify the lines i n the same pair: 5 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 Fig ur e 11 . Th e V - shaped pa irs . Key Poi nts Loc ation The location of the key points can be calculated with utilizing the V - shaped pai rs obtai ned f rom th e previous s tage. The algorithm can be explained as below: a) Extend the line s o f the V - shap ed p air u ntil t hey intersect each other b) Create the new center lines of an inter - finger space which has the bisecto r of an gle formed by the two lin es of th e corres ponding V - shape pair. c) The key poin ts f or palm print align ment can be located by calculating the intersecting points between the bo tto m curves o f vall eys and the ce nter l ines o f the correspon ding inte r - finger spaces. Ho wever the ab ove a lgor ithm c anno t w o rk correctly in some images due to the exceptional cases in the v - shape pai rs. As de monstra ted i n Figur e 12 , t he exceptional cases occurred usual ly whe n the i nter - space between the f inger does not wide enou gh. Different person has different finger shape, when people close their figures tig htly, the ga p sha pe in - between two figures can be different. It can be eit her a parallel V - sha pe line pair wi t h the sa me angl es’ val ue or an intersect ed V - shape pair intersected above the expected key point . To overcom e this issue, addi tiona l check ing is added to chec k which categor y the pair belongs to. Once it is confirmed, proper action will be taken to set the center line in above s tep b) as foll ows: - If it is a parallel li ne pair, the center line is set to be the middle line o f these two pa rallel lines. - If it is an i ntersected line pair, the center line is set to be the line which divides th e intersected angle into two halve s. Fig ure 12 . Excepti onal c ases o ccurred usu ally wh en t he int er - sp ace betw ee n the f ing er doe s not w ide eno ug h. Select ing Main K ey points When t he number of ke y po ints identi fied is more t han 3 , it means that the ke y point betw een thum b and in dex fing er is also detected. The system mus t exc lude it sinc e the ke y poi nts needed to locate the ROI are only K1, K2, and K3. I t is necessary to select three desired key points out of the four. Among the four points, ev ery three adjacent points can f orm a triangle. T here are two such trian gles in total. T his is illustrated in Fi gure 13. Fig ur e 13 . T ri ang le s F orme d by Fo ur Key Po ints . It is noticed that the triangle c ontaining K4 has a greater height than the other triangle, because K4 is relatively far 6 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 away from the oth er points. Hence, th e desired key points can be deter mined by selecting the triangle with smaller height. Step 4: E stablish the ROI Windo w Aft er the des ired key points (K1, K2, and K3) a re locate d, the coordinate sys tem can be established in order t o create th e ROI of the palm using t he way des cribed in s ecti on 4 . The size of ROI is dynamically determined by the distance between K1 and K3. It mak es the ROI extraction scale inv ariant. The distance between the camera and palm does not affect the range of the region we extract. It gives the user max imum freedom. A fe w sa mples o f RO Is extr acted from palms with flexible po sitions are shown in Figure 14. Despite the scale and gesture, the consi stent regions are extracted. Fig ur e 14 . ROIs extr act ed wit h flexib le pa lm p osition s. VI. L INE E XTR AC T ION Principal lines a nd co urs e wrinkle s ar e more obvi ous t han original ROI for huma n vis ion to j udge t he e fficie nc y o f RO I extraction. Line infor mation is highli ghted b y following step s [19 ]: i. Appl y the ave ra ging ma sk ii. Applying the line d etection masks iii. Thr eshold the i mage iv. Line t hinni ng Figure 15 illu strates the ab ove steps thro ugh the first exa mple in F igur e 14 . The line str uctur e is hi ghli ghted tha t it is much easier for human visi on to tell. Fig ur e 15 . Sa mple o f lin e extr actio n . VII. E XPER I MENT A ND R E SULTS As the main aim of this research is to im plement RST invariant ROI e xtraction method that is capabl e of handl ing pal m print i mages t hat ar e var ying in t erms o f fing ers positioning and distance to camera . A ne w m obile p alm prin t d ata base with one thous and f ive hundre ds phot os is f ormed. In this research, palm print images are captured using th ree mobile embedded cameras w ith different resolutions f rom two differ ent mobi le pho nes. Dur ing t he i mage capt uring pro cess, no fixed pegs were used to restrict t he mov ement, rotatio n and stretching of th e hands. Each device is us ed to capture images of bo th hand s fro m thir ty sub ject s. Fig ure 1 6 , 1 7 , a nd 1 8 sho w the sam ple images captured by different mobile phones. With each hand, fiv e photos are taken for each of the five different positioning of the hands in or der to test the r obust ness o f the algorithm. Fig ur e 16 . Pal m pri nt c apt ure d usi ng D81 0 VG A came ra . 7 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 Fig ur e 17. Palm pr in t cap tur ed us in g D 810 2MP c ame ra . Fig ur e 18. Palm pr in t cap tur ed us in g S G H - i9 00 5M P ca mera . The complete procedure to ext ract ROI in this research is illus trate d by one example show n in Figu re 19. Fig ure 19 . The c omplete p roced ure t o extra ct RO I in thi s resea rch . Figure 20 . T he images used for ex periment with v ariou s p osi ti onin g o f fin ge rs. Figure 21 . T he ext racted ROIs of palm pri nt s in Fi gu re 20 . To test the rob ustness of the new algo rithm, 15 palm images fro m one perso n used in the experiment. There is no constrai nt on pa lm posit ion and o rientation. T he subje ct can put her palm freely while taki ng the pictures. As for gesture, t hes e p alm im ages can be cl assif ied in to 3 classe s according to the abnor mality’s level of both t he palm’s shape and po sition. The first class c o ns ist s of 5 nor mal - shaped palm images with optimal in - space d ista nce bet ween the fi ngers . The secon d class consist s of 5 norm al - shaped pa lm images with various posi ti ons and i n - spa ce distance betw een t he f inge rs. The l ast class co nsist s of 5 abnorm al - shap ed palm images with va rious positio n and in - space dist ance betw een the fingers . Figure 20 shows the imag es used in the ex perim ent. Images in the s am e row re present th e sam e class accord ing to the 8 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) Inter natio nal J ournal of Compute r Sc ience and I nform atio n Secur ity, V ol. 6, No. 2, 20 09 sequen ce of class . E.g ., t h e im ages in th e firs t row are cl assif ied in to fir st cla ss, a nd so o n. The resu lts of all im ages are ex pected to be a pproxi mately the sam e, s ince th ey a re t aken from the s ame pe rson’ s palm . Figure 21 shows the result of ROI extracti on of the im ages in Fi gure 20 . O bserve f rom Figure 21 , the new tec hnique is approv ed to be succes sfu l as th e ROI s are ac ceptab ly consi stent. Take im age pair 11 and 14 as ex ample. Th e imag es’ scales , locat ions, and orientations vary a lot, the extract two ROI s are acce ptab ly co nsist ent. It shows that the new technique can tolerat e RS T vari ance wel l. Amon g the th ree g estur e cate gories , the ROIs in sam e row are v ery s im ilar t o e ach other. R OI s in s econd c las s are mo re similar to th ose in first class than th ird class. Although th e resul t s in th ird cla ss are not per fect , they are still sa tisfact ory sinc e they are consistent ly re presenting th e sim ilar significant palm print feature s as other tw o class es . VIII. C ONCLUS ION The im proved s quare - based palm print ROI ex tr action meth od was succ essfu lly im plemented and int egrated in to the current appl ication suite. In compariso n to other ROI e xtr actio n methods that are based on bound ary tracking of the overall hand shap e w ith limitatio n of being unable to pr o cess p alm print image s that have one or m ore f ing ers cl ose d, t he s ys tem can n ow ef fect ively han dle t he s egm entation of palm pri nt images with varying finger posi tioning. It opens up the possibility of bring ing the palm p rint techno logy mob ile. T hrough the exper iments and findings in this re sear ch, it was found tha t the images captured with latest mod el of t he mobile camer a at f ive m ega pixels , th e ver ifica tion rate w as close and co mparable to those of the palm print images i n the researc h benchmark. Thus, with the certain co ntinuous advan cem ent in m obile cam er a techn ologi es, it w ould be in n o time that the us age of palm print authenticati on in mobile sy stem s be prov ed as v iabl e an d prac tic al f or w idesp rea d app lic atio ns. Howev er, researche r s have proved that the identif ication perform ance o f m ost of th e u nim oda l m eth ods are no t satisfactory due t o a v ariety of probl ems s uch as n oise in da ta, restri cted deg rees of fr eedom , non - universality, and lo w accura cy . Fusing s everal mod alities together might be able to make system mor e robust [20] . In future, w e w ill investiga te a feat ure to work together with p alm print with out comprom ising the key ad vantage of our c urrent system, that is , high degree o f fre edom . R EFERENCES [1] B iometri c techn ologies - an introduc tion . Inte ll igence , Acuity Mar ket. 2007, B iomet ric Techn ology Today, p. 9. [2] A. Kong, D . Zhang an d G . L u. A Stu dy o f Ide ntical T w ins’ Pal m prin ts. s.l. : Sprin ger - Verla g Berli n Hei delb erg, 2 005. [3] T. Savič, N . Paveš ić , Personal recogn ition bas ed on an image of t he palmar surfa ce of the hand . Pat tern Recognition , Vol. 4 0 , Is s ue 11 , P 3152 - 31 63 , November 2 007. [4] W.X. L i, D. Z hang and Z .Q. X u. I mage alignme nt ba sed o n invar iant featu res for palm p rint id entific ation. Si gnal Process ing: Ima ge Commun ication. 2003, Vol. 1 8, [5] H. K. Po ly tec hnic Un ive rsity . Pal m print data base, 20 05. Biom etric Resea rch Cen ter Webs ite. [On line] 2005 . http: //w ww 4.co mp.po ly u.e du.hk /~b io metr ics/ . [6] W. X. Li , Authenti cati ng Pe rso nal I dentites Usi ng Pal m Pr int Recogni tion. s. l. : Hong Kong Pol ytechni c Universit y, 2004. [7] D. Zha ng, P almp rint Data . Palmpri nt Auth entica tion. s .l. : Sprin ger Scie nce + Bus ine ss Medi a I nc., 200 4. [8] A. Kumar, D . Wong , H. She n, A . Ja in , Perso nal V erif icatio n us ing Palm prin t and Hand Geometr y Biom etri c. Lect ure N otes in C omput er Science , vol.26 88, 2003. [9] C. Ha na; H. Cheng b, C . Linb, K . Fanb , Per sonal authe nti catio n usi ng palm - print features . Pa ttern Recogn ition, Vol. 36, P371 - 38 1, 2003. [10] A. R os s, A. Jain, I nfo rmatio n fus ion in biome tric s Pat tern Re cog nitio n Letter s vol. 24 , P 2115 – 2125 , 2003 . [11] C. Han , A ha nd -b ase d p ers onal aut hent icat ion usi ng a co ars e - to - fin e strate gy . I mage and Vi sio n Com put ing vol. 22 , P 909 – 918 , 2 004. [12] D. Zha ng, W . Ko ng, J. Yo u, M. Wo ng, O nline Pal m print I dentif ica tion . IEEE T ranac tions on P atte rn A naly sis a nd M achi ne I ntellig ence , vol . 25, no . 9, S eptemb er 2003 . [13] W. Kong, D . Z hang , W . Li , Pal mpr int featur e ex trac tion usin g 2 - D Gabor filters , Pa ttern Recognit ion vol. 36 , P 2339 – 2347 , 2003. [14] D. Zhang , V., N. L uo, A . Kumar , Ro bust pal m prin t ve rifica tio n us ing 2D a nd 3D fe atur es . Pat tern Recogn ition, vol. 43, I ssue 1, P 358 - 368 , Janua ry 2010 . [15] Hum an iden tif icati on in in forma tion s ystems : Man agem ent ch allen ges and publ ic po licy iss ue s. Cl arke , R . 1 994. [16] Zhe nhua G uo , Dav id Z hang , Le i Zha ng, W ang meng Zuo , Palmpr int ver ificatio n us ing bina ry orie ntat ion co - occurr ence ve ctor , Patt ern Reco gnit ion L etters , Vo lume 30, I ss ue 13, P 1219 - 1227 , Octo ber 200 9. [17] M. K. Leu ng and Y. H . Yang, “Dynami c Two - strip A lgo rithm in C urv e Fitting ”, P atte rn Re cogn itio n, Vol .23, pp.6 9 - 79, 1990. [18] X. W u, K. Wang , A no vel appro ach of p a lm - line ex tractio n, Pro cee dings of the th ird intern ational c onference on image and graph ics, 2004 . [19] R. C. G onzal ez and R. E. W oo ds, Dig ital I mage Proce ssing (S econd Editio n), P rent ice - H all I nc., Ne w J ersey , 2002. [20] J. W ang, W . Yau, A . Suwandy , E . Sung , Pers on re co gniti on by fusing palm prin t and pal m ve in im ag es ba sed o n “L apl acian palm ” repres ent ati on , Pattern Rec ogni tion , vo l. 41, I ss ue 5, P 1514 - 1527 , M ay 2008. 9 http://sites.google.com/site/ijcsis/ ISSN 1947-5500