face verification with age progression using discriminative method and gradient orientation pyramid

Upload: ijafrc

Post on 23-Feb-2018

221 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    1/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    + - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    Face Verification .ith Age 0rogression using 1iscriminative

    ethod and 3radient 4rientation 05ramid1Piyush D. Hegu, 2Pradnya J. Suryawanshi

    1Dept of E & TC, Assistant Professor,Jawahara Darda !nstitute of Engineering & Te"hnoogy, #a$at%a

    2Dept of Ee"troni"s, Assistant Professor, Priyadarshini Coege of Engineering, agpur1piyush.hegu'g%ai."o%, 2pradnya(s'rediff%ai."o%

    A 6 % 7 R A C 7

    Facial anal5sis .ith respect to age verification .hile considering several face related challenges is

    one of the open prolems in computer vision s5stem, as alread5 so man5 good algorithms e9ists

    for face verification .ith age progression$ 7he human face is the premier iometric in the field of

    human recognition ecause of its eas5 ac:uisition$ In this paper, .e propose a methodolog5 for

    real time facial image anal5sis and verification across age progression$ 7here are several

    challenges in face verification li;e different poses, e9pressions, ac;grounds and illumination

    conditions ecause of .hich the tas; ecomes difficult$ 7he facial image is processed using

    support vector machine .hich is an e9ample of discriminative model used in the machine

    learning$ 7he effective and simple representation can 8e o8tained 85 finding gradient

    orientation of a facial image, so first it needs to calculate 3radient orientation (34)$ 7his

    representation isimproved .hen image is represented in p5ramidal form, .hich results in

    the use of the gradient orientation p5ramid ( 340)$ 7his is then comined .ith a support vector

    machine (%V) to give e9cellent performance$ 7his .or; e9plains ho. age differences and the

    other factors such asimage :ualit5, spectacles, and facial hair affect recognition algorithms$

    Index Terms: Face verification, age progression, gradient orientation pyramid, support vector machine

    I$ I&7R41

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    2/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    +# - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    The reuire%ents of a good fa"e re"ognition agorith% are high re"ognition rates, toeran"e towards

    $arious en$iron%enta fa"tors su"h asiu%ination, fa"ia poses, fa"ia e/pressions, i%age *a"grounds,

    i%age s"aes, hu%an ageing and aso good "o%putationa and spa"e "o%pe/ity.

    II$ =I7>RA7 R>VI>?

    3a%anathan & Cheeppa 456proposed a 7ayesian age8differen"e "assifier that is *uit on a pro*a*iisti"

    eigenspa"es fra%ewor. 9hou & :eorges"u 41;6presented genera !73 -!%age *ased 3egression using

    *oosting to see"t ree$ant features fro% the i%age. Aso tested the proposed !73 agorith% on Age

    esti%ation, Pu%onary Tu%or dete"tion & endo"ardia wa o"ai+ation. 3a%anathan and Cheeppa 456

    use a fa"e growing %ode for fa"e $erifi"ation tas for peope under age of 1

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    3/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    +B - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    The tas of our study is two8fod. The first is to in$estigate representations and agorith%s for

    $erifi"ation. The se"ond is to study the effe"t of age differen"es and the other fa"tors su"h as i%age

    uaity, spe"ta"es, and fa"ia hairon $erifi"ation agorith%s. e use three datasets in our study. Two of

    the% are passport datasets in$o$ing %ore than 1, F4R

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    4/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    +* - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    where, / 3n is the feature $e"tor e/tra"ted fro% the i%age pair -I1, I2 through the feature

    e/tra"tion fun"tion F0 I ! I " 3n , I is the set of a i%ages, and 3n for%s the n8di%ensiona feature

    spa"e. Then SF is used to di$ide the feature spa"e into two "asses, one for intra8persona pairs and the

    other for e/tra8persona pairs. e denote the separating *oundary with the foowing euationNs

    - Iiyi> -Si,x * K L (2)i=1

    here, #s is the nu%*er of support $e"tors and si is the ithsupport $e"tor. L is used to trade off the

    "orre"t re(e"t rateand "orre"t a""ept rate as des"ri*ed in -@ and -M. $-%,% is the erne fun"tion that

    pro$ides SF with non8inear a*iities. !n our e/peri%ents, we use the ?i*SF i*rary. )or $erifi"ation

    tass, the "orre"t re(e"t rate -C33 and the "orre"t a""eptan"e rate -CA3 are two "riti"a "riteria,

    &'' = correctly reected extra personal pairs* total extra personal pairs (+)

    &' = correctly accepted intra personal pairs* total intra personal pairs (-)

    here Na""eptO indi"ates that the input i%age pair are fro% the sa%e su*(e"t and Nre(e"tO indi"ates the

    opposite. !n addition, the e.ual error rate -EE3, defined as the error rate when a soution has the sa%e

    CA3 and C33, is freuenty used to %easure $erifi"ation perfor%an"e.

    6$

    3radient 4rientation And 3radient 4rientation 05ramid

    :radient8*ased representations are re"enty widey used in "o%puter $ision and pattern re"ognition

    tass. e dis"ard gradient %agnitude infor%ation and use ony orientations, whi"h de%onstratessignifi"ant i%pro$e%ent in our e/peri%ents. )urther%ore, the gradient dire"tions at different s"aes are

    "o%*ined to %ae a hierar"hi"a representation.

    -a !%age I -* Pyra%id /-I -" :P -d 0-I

    Figure '$ Computation of a 340 from an input image I

    :i$en an i%age I-p, where p K -x y indi"ates pi/e o"ations, we first define the pyra%id of I as

    /-I K ! -pQ RsRK; with:

    I-pQ ; K I-p,

    I -pQ R K 4I-pQ R81 U -p6 V2 R K 1,....,s, -B

    here U-p is the :aussian erne, denotes the "on$oution operator, V2 denotes haf si+e

    downsa%ping, and s is the nu%*er of pyra%id ayers. ote that in -B the notation Iis used *oth for the

    origina i%age and the i%ages at different s"aes for "on$enien"e.

    Then, the gradient orientation at ea"h s"ae R isdefined *y its nor%ai+ed gradient $e"tors at ea"h pi/e.

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    5/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    +8 - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    g -I- pQ R K -! -p, RW X -! -p, RX if X-I-p, RX Y Z

    -;, ; Z otherwise -

    here, Z is a threshod for deaing with NfatO pi/es. The gradient orientation pyra%id -:P of I, is

    naturay defined as 0-I K sta"-g-I-p, sRK; G 3n2that %aps I to a d 3 2 representation, where sta"

    -% is used for sta"ing gradient orientations of a pi/es a"ross a s"aes and n is the tota nu%*er ofpi/es. )ig. @ iustrates the "o%putation of a :P fro% an input i%age.

    C$ @ernels 6et.een 340s

    :i$en an i%age pair -I1 I2 and "orresponding :Ps -01K 0-I1Q02K 0-I2, the feature $e"torxK F -I1

    I2 is "o%puted as the "osines of the differen"e *etween gradient orientations at a pi/es o$er s"aes.

    x = F -I1, I2= -:1 :2 -

    here isthe ee%ent8wise produ"t. e/t, we appy the :aussian erne to the e/tra"ted feature xto *e

    used with the SF fra%ewor. Spe"ifi"ay, our erne is defined as

    $ -/1, /2 K e/p -8[X/18 /2X2 -%

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    6/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    ++ - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    -a -*

    Figure B$ (a) Final matching Results of t.o images () Command .indo. sho.ing matching percentage

    !n the first phase of e/peri%enta resuts, we "o%pare the i%ages fro% ):net data*ase of sa%e person at

    two different ages -2M and years. )igure B shows feature %atri/ "o%ponents of two i%ages fro%

    whi"h the different intensity $aues "an *e o*ser$ed, whi"h "an *e used for "a"uating %at"hing

    per"entage. The fina resut showing the *est %at"hing with the see"ted fa"e i%age -)igure -*

    whereas figure -* shows the %at"hing per"entage with the other i%age.

    The age differen"e *etween two i%ages is so %u"h -i.e. M@ years *ut sti we got the %at"hing per"entage

    of a*out

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    7/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    "!! - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    -a -*

    Figure +$(a) final matching Results of t.o images () Command .indo. sho.ing matching percentage

    !n the se"ond phase of e/peri%enta resuts, we "o%pare the i%ages fro% ):net data*ase of sa%e person

    at different age groups -fro% age 18 M; years ha$ing age differen"es of @8< years.The fina resut showing

    the *est %at"hing with the see"ted fa"e i%age -)igure 5-a whereas figure 5-* shows the %at"hing

    per"entage with the other i%ages fro% this the *est %at"h is see"ted whi"h is indi"ated *y red *o/.

    -a -* -" -d -e

    Figure "!. Face images of same person having different face related changes ((a) &ormal Image (d) change

    of e9pressions, () presence of spectacle, (c) change of position, (e) change of image ac:uisition

    conditions)

    Figure ""0 Final matching result

    -a -*

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    8/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    "!" - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    Figure "20 (a) Command .indo. sho.ing total numer of entries in the dataase () Command .indo.sho.ing matching percentage .ith other images

    !n the third phase of e/peri%enta resuts, we "o%pare the i%ages of sa%e person ha$ing different fa"e

    reated "hanges -su"h as "hange of fa"ia e/pressions, "hange of i%age a"uisition "onditions, presen"e of

    spe"ta"es, and "hange of fa"ia position with the nor%a fa"e. So fro% this "o%parison we got the

    %at"hing per"entage of nor%a i%age with different i%ages whi"h "an *e shown in figure 12-*. The fina

    resut with *est %at"hing per"entage of 51.;

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    9/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    "!2 - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    e pan to in$estigate se$era dire"tions in our future wor. )irst of a, testing on a arge pu*i" dataset

    wi *e "ondu"ted for deeper understanding of the proposed approa"hes. e pan to wor on the 3PH

    dataset for this purpose. Se"ond, we pan to appy other dis"ri%inati$e approa"hes -e.g., *oosting for

    si%utaneous feature anaysis and "assifi"ation. The %ost i%portant is to "reate and anay+e our own

    data*ase for% our region so as to "ontri*ute one step ahead towards the resear"h in fa"ia $erifi"ation

    and re"ognition te"hnoogy.

    ID$ R>F>R>&C>%

    416 S. 7iswas, :. Aggarwa, . 3a%anathan, and 3. Cheappa, NA nongenerati$e approa"h for fa"e

    re"ognition a"ross aging,O 4iometrics: Theory, pplications and 5ystems, 2667% 4T5 2667% 2nd

    I888 International&onference on, pp. 1], 25 2;;

  • 7/24/2019 Face Verification with Age Progression using Discriminative Method and Gradient Orientation Pyramid

    10/10

    International Journal of Advance Foundation and Research in Computer (IJAFRC)

    Volume 2, Issue 8, August 2!"#$I%%& 2'8 8#', Impact Factor "$'"*

    "!' - 2!"#, IJAFRC All Rights Reserved ...$i/afrc$org

    A

    0i5ush 1$ egu was *orn in #a$at%a, aharashtra state, !ndia, in 15