• 沒有找到結果。

類神經網路於生物測定認證技術及應用之研究---子計畫I:類神經網路靜態人像辨識系統之研究(III)

N/A
N/A
Protected

Academic year: 2021

Share "類神經網路於生物測定認證技術及應用之研究---子計畫I:類神經網路靜態人像辨識系統之研究(III)"

Copied!
4
0
0

加載中.... (立即查看全文)

全文

(1)

(2)  

(3)  The Study of Neural Network Face Recognition System for Static Images (III)  .

(4)

(5)         .       !"#$%&'()*+,    

(6)  

(7)   !" #$%&'("

(8)  )*+,-!./0123456 7 89:

(9)  (" #$ ;%<=>4 ?@ABCD:. EFGHI89:

(10) BJ( KLMNOPQ RRSTU VWX&'("

(11)  0<YZ> [\]^_`abcdeefg

(12) h 45 6ijklmn'o pq:(" 

(13) rstuvp(wxyz {4 ?@Aj|} ~~

(14) 

(15) rs tB>/D€% ‚ƒ„‚ a9:

(16)  „†† : f<=p ‡ˆƒ‰ a9:

(17)  ~††† : f) Š‹Œ

(18) r stxy/D€% ‡„ƒ a9:

(19)  ~††ˆ : f<=p ‡ƒ a9:

(20)  ††† : f4Ž‘ ’Uh“”l•–—˜™š›œhttp:// www.csie.nctu.edu.tw /~nnlabž4   œ

(21) Ÿ ]^_`Ÿ  ŸZ>[\]^_`  Abstract. to non-uniform and yet to achieve better recognition performance. In this project, we present a PDBNN based feature reduction algorithm that deletes some feature vectors which contribute the least among of the whole feature set. The deletion is performed on individual facial basis. By applying the proposed algorithm, we performed some face recognition experiments on a in-house 151 people facial database and the ORL database. The experimental results show the recognition accuracy improved from the original 86.26% (2500 features) to 94.87% by using only 1000 features on the in-house database, and the recognition accuracy improved from 92.5% (10304 features) to 98.5% (8000 features) on the ORL database. Keywords: Face Recognition, Neural Networks, Feature selection, PDBNN . 

(22)  

(23)   ¡ ¢ £ ¤ ¥ a¦§¨¨©ª« ª©¬­®«¯¨¯­«f°o D£

(24) ( "±²³´µ¶h·j¸¹+º W!»¼²½¾)¿À8DÁ ÃÄ'

(25) ÅƲ4ÇÈÉÊË% –WÌ%ÌÍÎÏÐp

(26) Ñ ÒÓÔÕ)֛*×ØÙŽ WÚ°("ÛÜ1<m%ÝÞÊ ß a®©­à©¨ª¯¬§á⩧¨ãª©äf å a©¯®©«â§¬©äf  æ ç è é a¨©à¦á§¨© ২¬ê¯«®f¤”èéa®ª§¦ê২¬ê¯«®f   ] ^ _ ` a«©ãª§á «©¨ë­ªìäf  0 í. Selecting proper and important features is an essential task for efficient face recognition. Most of recognition or classification algorithms proposed to extract and to use features in uniform manner from each object. We believe the constraint on uniform number of features could be relaxed 1.

(27)  #$4 56–klmûB ÛÜW 0·YZœ  Jn¤”Û ÜjW ¬á§ää u !" [ Bš8n \]ô   

(28) FG^D=S²rs [ Bm_9: ¬á§ää  uU`v!»SabS²[ ó 5jW ¬á§ää c !" bS² [ Bš84 d nef 

(29)   

(30) 

(31)  

(32)   gcTBJ° 4*+J hi £jÕm  [ jUSa ¬á§ää  klWmÊ: ¬á§ää  l S²nbo¼Bpq)*Ê: ¬á§ää ^\cr˜û>nb  %s*Ê: ¬á§ää I8?L r s L M  wÈt^\uv' ¬á§ää rswx˜*Ê: pq  ¬á§ää4Wy£z56klm: Um9: ¬á§ää { [ F G0íU (" [ Bš8n  )4  )×|ÔW (" [ Bš8n U567 8 Z > \ ] ^ _ ` f       

(33)       g }]k &' Í ¤Ÿ

(34) û&'¤ ~68ô2œ bcdee  W

(35)  

(36) 

(37)  kjW    ¤B

(38) å&.Em 7 !" #$ 56€

(39) å:; <=h("r>"?0

(40)  

(41) 

(42) 

(43) %9 ô@A6

(44) åýBCòr> : ¬á§ää jýo 

(45)   8ÚÃ DEF/ BGm

(46) åj HI { ‚› )ƒjW 

(47)  „ ó5š8åJ&.1í0²234( ! "  [ 456j  Û4) JK9:

(48) u/ 07 ?LMNrs@GmO

(49) j W

(50) PüQRù0í:'

(51) M Nrs½¾¼²i9: S² *q:Q!ITQÌ6•O ý £O

(52) Uh6Vi9:

(53)  gWXm:'

(54) B î¯ïï©«ð§ªì­ñð­ï©áä òò4ó)ô– õz{öx ÷Wøùöúû üýjU

(55) rst8þ ,-jUú

(56) rst

(57) ³´±²µ¶,-ò4

(58) z{I8E £ Èj7 

(59) :  > 

(60) :þ 4ßm

(61) 0 šû>ÚŽ ó)pŽ÷óWþ U p=>1<m4?@ ŽI8klm:Wû

(62) . !EÈ" j<m û#$z$—˜W=>%

(63)

(64) û4       ?@Aû&'D(۟  Ÿ )ò*:æ+'4(Û  ,-

(65) å&.Em / 0² 1 2344 !E,-ý2345

(66) m

(67) å ))6 7 jEm

(68) š849 :û&'¤j•4 . 2.

(69)  )&'¤ j•. %­pÌ%ÌÏÐ)Oª « ¬ ÃÄ4ûhr>®¯í°± UWúx§o 4?@Aj|'h  ]^_`

(70) û/g Oª «¬ûJn4zû  ²³´Û µ¶ÃÄ·¸Z¨¹ Wºh»¨Ñ4  . ¤ )&'  xyz{×56ijklm n'o pq:("

(71) rst uvp(wxyz{4 ?@Aj |} ~~

(72) 

(73) rstB>/ D€% ‚ƒ„‚ a9:

(74)  „†† :. f<=p ‡ˆƒ‰ a9:

(75)  ~††† :. f) Š‹Œ

(76) rstxy/ D€% ‡„ƒ a9:

(77)  ~††ˆ :. f<=p ‡ƒ a9:

(78)  ††† :. f4D£9:

(79) ü 8JUO

(80)  g†‡¿/ (" [ Bg

(81) )ˆ @‰9:

(82)  FmkF. U)/0˜€X{4    ?@A˜{/n*×%Š‹Œ ~ƒ ?@A8kl퍎

(83) û h Ô  | ' 8”

(84) rs t

(85) ‘o ~†††

(86) / Ž6 ‘o’E“BǔJ

(87) r stz

(88) rstI/g?@Ax yÈ !• 36– 5—/ <Yǔ!•ØÙÍÎ ˜m™ |-

(89) rst

(90) š›4 „ƒ :

(91) û>=b

(92) œûa   

(93) fXUWþ •¡4

(94) œœ˜’> œjÔ

(95) ˜žX{) ô7 Ÿ %œ

(96) û€  ¡ o £%

(97) <=aª©§á¨¯à©f B456

(98) œ7 8:;œ ¢"?0ô£;6a

(99) åýBCf% H

(100) (Rœ˜ ’>=œjÔ

(101) !»^Ô ¤q¥/o £%

(102)

(103) ¦û B4?@Aj—˜

(104) œ

(105)  ûhx§Á¨א¤*j •4 ƒ o Á©££Oª «¬ûÉ. \2^!T/!Blbnbutv-!U/!Tbtblj-!I/!Gvlbnbdij-!boe! Z/! Tvfobhb/“B! spcvtu! gbdf! jefoujgjdbujpo! tdifnf.LM!fyqbotjpo!pg!bo!jowbsjbou!gfbuvsf! tqbdf-”jo! TQJF! Qspd/;Joufmm/! Spcput! boe! Dpnqvufs! Wjtjpo! Y;! Bmhpsjuint! boe! Ufdi/-! 2718;82.95-!2::2/! \3^!S/!Csvofmmj!boe!U/!Qphhjp/!“Gbdf!sfdphojujpo;! Gfbuvsft!wfstvt!ufnqmbuft/”!JFFF!Usbotbdujpot! po! Qbuufso! Bobmztjt! boe! Nbdijof! Joufmmjhfodf-26)21*;2153..2163-!Pdupcfs!2::4/! \4^! J/K/! Dpy-! K/!Hipto-!boe!Q/O/!Zjbojmpt/! “Gfbuvsf.cbtfe! gbdf! sfdphojujpo! vtjoh! njyuvsf.ejtubodf/”! Jo! Dpnqvufs! Wjtjpo! boe! Qbuufso!Sfdphojujpo/JFFF!Qsftt-!2::7/! \5^!U/!Lbobef/!Qjduvsf!Qspdfttjoh!cz!Dpnqvufs! Dpnqmfy!boe!Sfdphojujpo!pg!Ivnbo!Gbdft/!Qi/E! uiftjt-!Lvzup!Vojwfstjuz-!2:84/! \6^!N/!Mbeft-!K/!Wpscsvhhfo!-!K/!Cvinboo-!K/! Mbohf-! boe! D/W/E/! Nbmtcvsh/! “Ejtupsujpo! jowbsjbou!pckfdu!sfdphojujpo!jo!ezobnjd!mjol! bsdijufduvsf-”! JFFF! Usbot/! Dpnqvufst-! 53;411.422-2::4/! \7^!T/!Mbxsfodf-!D/M/!Gjmft-!B/D/!Utpj-!boe!E/! Cbdl/! “Gbdf! Sfdphojujpo;! B! Dpowpmvujpobm! Ofvsbm!Ofuxpsl!Bqqspbdi/”!JFFF!Usbotbdujpot!po! Ofvsbm!Ofuxpslt-!Wpm/9-!Op/2-!Kbovbsz-!2::8/! \8^!Tiboh.Ivoh!Mjo-!T/Z/!Lvoh-!boe!Mpoh.Kj!Mjo/! “Gbdf!Sfdphojujpo0Efufdujpo!cz!Qspcbcjmjtujd! Efdjtjpo.Cbtfe! Ofvsbm! Ofuxpsl-”! JFFF! Usbotbdujpot!po!Ofvsbm!Ofuxpslt-!Wpm/9-!Op/2-! Kbovbsz-!qq/225..243-!2::8/! \9^!B/!Qfoumboe-!C/!Nphibeebn-!U/!Tubsofs-!boe! N/!Uvsl-!“Wjfx.cbtfe!boe!npevmbs!fjhfotqbdft! gps!gbdf!sfdphojujpo-”!jo!Qspd/!JFFF!Dpnqvufs! Tpd/! Dpog/! Po! Dpnqvufs! Wjtjpo! boe! Qbuufso! Sfdphojujpo-qbhf!95.:2-!2::5/! \:^!Ifosz!B/!Spxmfz-!Tivnffu!Cbmvkb-!boe!Ublfp! Lbobef-“Ivnbo!gbdf!efufdujpo!jo!wjtvbm!tdfoft-”! Ufdiojdbm! Sfqpsu! DNV.DT.:6269-Tdippm! pg! 3.

(106) Dpnqvufs!Tdjfodf-!Dfsofhjf!Nfmmpo!Vojwfstjuz-! !!!!!!¤*Ÿûx§¨×! Qjuutcvshi-!QB-!Kvmz!2::6/! \21^! G/T/! Tbnbsjb! boe! B/D/! Ibsufs-! “Qbsbnfufsjtbujpo!pg!b!tupdibtujd!npefm!gps! ivnbo!gbdf!jefoujgjdbujpo-”!Jo!Qspdffejoht!pg! uif! 3oe! JFFF! xpsltipq! po! Bqqmjdbujpot! pg! Dpnqvufs!Wjtjpo-Tbsbtpub-!Gmpsjeb-!2::5/! \22^! G/T/! Tbnbsjb-! “Gbdf! Sfdphojujpo!vtjoh! Ijeefo! Nbslpw!Npefmt-”!Qi/E!uiftjt-!Usjojuz! Dpmmfhf-!Vojwfstjuz!pg!Dbncsjehf-!Dbncsjehf-! 2::5/! \23^! M/K/! Tifo-! I/D/! Gv-! boe! Z/Z/! Yv-! “B! qsjodjqbm!dpnqpofou!cbtfe!qspcbcjmjtujd!ECOO! gps!gbdf!sfdphojujpo-”!jo!JDJQ(:7-!Wpm/4-!qbhft! 5::..614/! \24^!Lbi.Lbz!Tvoh!boe!Upnbtp!Qphhjp-!“Mfbsojoh! ivnbo!gbdf!efufdujpo!jo!dmvuufsfe!tdfoft-”!Jo! Dpnqvufs!Bobmztjt!pg!Jnbhft!boe!Qbuufsot-!qbhft! 543..54:-2::6/! \25^!N/!Uvsl!boe!B/!Qfoumboe-!“Fjhfogbdft!gps! sfdphojujpo-”! K/! pg!Dphojujwf!Ofvsptdjfodf-! 4;82..97-2::2/! \26^!M/K/Tifo!boe!I/D/Gv-!“Gfbuvsf!Sfevdujpo! gps!Gbdf!Sfdphojujpo!cz!QECOO-”!jo!JDT’:9/! \27^! Wjd! Csfoobo! boe! Kptf! Qsjodjqf-! “Gbdf! Dmbttjgjdbujpo! vtjoh! b! Nvmujsftpmvujpo! Qsjodmqbm!Dpnqpofou!Bobmztjt”-!jo!OOTQ’:9/! !. 4.

(107)

參考文獻

相關文件

電機工程學系暨研究所( EE ) 光電工程學研究所(GIPO) 電信工程學研究所(GICE) 電子工程學研究所(GIEE) 資訊工程學系暨研究所(CS IE )

Tekalp, “Frontal-View Face Detection and Facial Feature Extraction Using Color, Shape and Symmetry Based Cost Functions,” Pattern Recognition Letters, vol.. Fujibayashi,

倒傳遞神經網路的演算法使 SPOT 假色影像轉換到 SPOT 自然色影 像。影像的結果。(3)以不同天的 SPOT 假色影像進行網路回想,產 生

蔣松原,1998,應用 應用 應用 應用模糊理論 模糊理論 模糊理論

The purpose of this paper is to achieve the recognition of guide routes by the neural network, which integrates the approaches of color space conversion, image binary,

F., “A neural network structure for vector quantizers”, IEEE International Sympoisum, Vol. et al., “Error surfaces for multi-layer perceptrons”, IEEE Transactions on

通常在研究賽格威這類之平衡系統時在於機構之設計是十分的昂貴,本論文

本研究以河川生態工法為案例探討對象,應用自行開發設計之網