B!tuvez!po!uif!ofvsbm!ofuxpslt!gps!jnbhf!dpnqsfttjpo!
!
!"#$%&'($) *+
,-./
/0012300451617408749:
;-< ==>
!" # $ %&'()*+,-./01 2345 6 789:;<=> ?@A#; BCD EFGAHIJ&'(K LM2NO#P@Q+RS HIJ 34 TUVWXY Z[\Y]8 Z[VWXY \Y]8 2^_`abcdZ[R efghijkklmfZ[nXVopqrZ[ "#stuv!2HIJ34wRxuS 2yz{|2}~)HIJ R 2"#; _(,$HIJ&' (2yzEFTU-2 \o> HIJZ[`- XVopq mf Z[#EF2AHIJ2}~ AHIJ ¡`AHIJ ¢ £¤_¥#; ; \; ¦§; ; ¨©; ¨§ª¨ª«¬; ¨¨ª®; ®«¯°§ª«¬; ±«§ª¨©¯; ª; ®«¬®§¬²; ª¯±; ®«¯°§ª«¬; ª; ; ³§´; ¯¨µ§; ¨®©¬ª¶µ·; ; >©§; ¸ª¨; ¹º¬«¹¬; ±«§ª¨©¯²; ¯«¬±;¨©¯²;789:;<=>;ª¬±;¨©;¯«¨;¦§¶µ¬¨´;µ; ¯¨©«·; V«;¦§²;¯¬´;¨;®«¯°§ª«¬;±«§ª¨©¯;;«¬; ¬µ§;¬¨¹«§º;©³;¬;°§«°«²;·±·²;VWXY;¯«²; \Y]8; ¯«²; VWXY; ¬; \Y]8; ©´§ª; ¯«; ¬; XVopq; ¯«·; ; »«¹³§²; ¨©; ¸ª¨ª¬±; ¬µ§; ª¯±; ®«¯°§ª«¬;¨ª;©³;¯¬´;§¹®º·; ¼¬; ¨©ª; °§«½®¨²; ; ¬¹; \o>; ª¯±; ®«¯°§ª«¬; ¬; ¬; ª¯°§«³; XVopq; ¯«; §; °§«°«·; ; ¾ª¬±; ¨©; ¯¨©«²;¨©; ¨§ª¬ª¬±; ¨ª¯; ®¬; ; ®§·; ; ¼¬; ª¨ª«¬²; ¨©´; ¯º; ¨©;®«¬¨§µ®¨ª«¬; «¦;¬µ§;¬¨¹«§º;ª§²;¬;¨©;µ;«¦;¬µ§;¬¨¹«§º; «¬;ª¯±;®«¯°§ª«¬;¯«§;°§®¨ª®·; ; HIJ¿,ÀÁÂÃÄÅ2 Æ ÇÈ %2&'(RxÉÊ2ËÌ#ÇÍ % &'(£RÎb2Ï¥ÐnÑrÒÓ ÔÕÖ×nØr ÒÖ^_ÏÙ×nÚr£ÛÜÝÞ"#ßP@Q t- ¢+R 2HIJ ÛÜ(à TU¿+ Pááâ2 ãäå u æv./çè2éê#; HIJ£Rëìíî2ÝÞ#ëìíîï ,ðÊñò óUô6õötuò2÷R Ïøù6EFú !tÊñ2ò)û üÎÇF2Ïøý¤#5þEF¤6vëìí îÈò,RÌ2éê¿ï,HIJ B£Rò2ÝÞù6 F AHIJ34K t-NO#tÁÂÆ +Rú2HIJ & ' ( T U µ¨««®ª¨ª³; ¯µ¨ª´§;°§®°¨§«¬;\Y]8;ÑÚ¦«§¬±ªª¬±; ¦¨µ§;¯°;VWXY;Ñ\Y]8 VWXY ^_c dZ[Ñ\Y]8 VWXY abcdZ[ÑR efghijkklmfZ[XVopq;Ø"#; 6¢ùT2-+v!2HIJ3 4 6RefghijkklmfZ[2é Ý?Ç2éÝÝ { 789:\Y]8VWXY cdZ[2éÝÍ Æ\Y]8 Z[ Æ ) VWXY 2éÝ,?Ë#stwRÎ2yz 6 Ð; !";}~{#./$®º°§«°±¨ª«¬ Û Ü(à` ¦«§±¬ª¨ª«¬ ÛÜ(à%&' ()}~ßP*P+@Q#; ,";-./HIJ012 -012IJ3-4«©«¬¬ 2IJ3 -"#; 5"; 2-./ÛÜ67 ¦«§±¬ª¨ª«¬ ÛÜ(à 2 ¬ª±©«§ Ê8` ¬ª±©«§ Ê89267":#; ;";éêw <Ñ#; R=>¢?HIJ&'(2yz t-CD EF$tuyz$@ÎABx /CÇFHI Jþ ¢2DE#EFTUF-6 \o> HIJ@2 Z[ÐnÑr\o> HIJZ[ `nØrhi \o> HIJZ[#G EFH$Refghi jkklmf Z[ nXVopqrIZ[Êk}~ 2yz#; 6ÎGJKLMEF2HIJZ [#; !"#$%&' t-HIJZ[N, fuzzy ARTMAP2[HIJÆ¢Ç,-O|[ HIJ¿ïPÇQRx)STUV AWúX.YCZ[\/2 ` Ê 8 # ð 2 ]z K ^ t -ART-based HIJZ[,_>`H 34# ART-basedHIJZ[ a£R STd\HIJ2bÝÆÇH£RÎ b2cLÐn1rt- EFd -iefinvigilance parameterr#5Ç2 ghQR--iAWjk×n2r t-Z[2}~&'(,lmn62#EF2&' (}~ È}~ ùoP2éê,ph. À2ù6EFù2}~ phqrp hn6# t-HIJZ[ EF`0st ú2Z[Ðn1rART-based HIJZ[ `(2)hi ART-based HIJZ[(u! ùv)#tst ART-based HIJZ[H IJ¢,wx.)}~`yì{#RÎ búÆLÐn1rFW zHIJ2 zâB úW,Iò){W,J{LM2 hi×n2r{2òúW|} ~`~){WàW 2^ iC?8i×n3r{W G ^iC?8i )WàúrT# ( )*+, t- ART-based HIJ,- 22c d[HIJÇ|} z2âB2 UnC outstarr2uùv5 z2 âB20- ART-based HIJ) U 2à,- outstar HIJ# z2û zjk)ßGt-jk PâB2 # âB2 !ð}~âB íîU âB#âB2 2IJ3-,x}~2 {#)O )tuIJ3 á !x -âBù6EFtuIJ3@âB IJ3#GEF¿¤GâB2@-MAXNETq@$>- zâB%ûQR -âBIJ3Q Ý)5Ç2IJ 3àúT#$-IJ3)*%R >IJ3 z22EFtt @`HÇN, zâB` H2# âB2 EFH-ef K}~{#t-i2Ê8GQ P U2Ê8`2éêÍÊ2efiQ UÍ82HIJs.$¡0ÍË2 ¢Ù×&ÆàQUÍÊ2HIJÈÍ 2¢Ù# U2à,-~£Ç$ùR2â BIJ3~0?32F3~)t-?8 3ýõö U2 ùIJ32-#-âBIJ32F3~, !Ç U22 ùRIJ3Æ2 #i.ù?- zâBûQÝâB2 2-âB IJ3ãäGÇù_2F3~i U2 U ¤t- U2F3~ý, zâB2~# -( ./0!"12 t- ART-based HIJZ[2}~& '(,-F¥2cd}~&'(¦-} ~§¥ EFA-p ART }~&'( K}~ ART HIJ×)¦F-}~§¥ E F A counter-propagation } ~ & ' ( 2 GrossbergÛÜ(àK}~ U2# 3( 4506!"12 G¨ EF-G ~âB~`^iC?8i nr©,hi ART-based HIJZ[ ªRr «P ¬#.EFG~ ®¯âB~ ãîU5.$2â B$> ART-based HIJZ[)*t
-âBý, {2ò×s$>hi ART-basedHIJZ[)*ÇH -°2±)²ï,G^iC³8i P .ù$®U2âB # (789:;<=>?@A; #$ ðRefghijkklmfHIJZ [nFSRVQr EFë´oP-[éê c2HIJ[µ¶Êk2} ~t-HIJ¢#)äEFûÝA·¸ ¹B(KóP-Íd\2HIJ²º ù2}~à,0»0O ttvQ¡ 0Çþ¥¢2¼½# @ /}~2NOEF$5u 2#)EF2Z[ EF6-8 2HIJ[KîõIR2²-ÊHIJ [ãäGIZ[ 2Refg`H ¾¿>HIJ[Æ-`Hfg%£ RS2HIJ[5u 2 ùv# EFÆù6Qr25IqÎÐn1r $>}~âB)tÇFvû¤}~ÀÍ8 2HIJù2}~GQÊÁ 8×n2ri>EFGIÊ2HIJ[ `0-8HIJ[EF¤6Gtu 82HIJ`¯ ÃÄÅ¢¤ } ~tAoEF2Z[ù2}~ Æ8 >IZ[×n3r$>/HIJ) *EF2(w,A·¸¹B(s,$>8 2HIJÊ2HIJÍÔ ¡[Ud\2HIJ×n4rEF2Z [ áÇ2HIJ[%ûLMåu Ïø.À2 zjkEF¤6Í82[K Y[U[âB# FSRVQ Z[,- §¦ 2@Refg`H¦F2,-8H IJ[)ÎK22à,hijk ~#Z[ÈIZ[.hijk~ 22`~2Ê8,iAWS2È C~ÈÊàéêÈs¿¡0ù2 ~3O # t- FSRVQ Z[ EFÉ - ¥ 2 c d [ Û Ü & ' ( 5 | } backpropagationÛÜ(àGEF¿G.ù TU2 ART-based Z[ 2ÛÜ&'(| } #¦-}~§¥ EF.A {2 ART ÛÜ(àK}~¦2 2Re fg`H#¦F-}~§¥ EFàA backpropagationÛÜ(àK}~¦F2 2HIJ[À#Êx2§¥ E FËÖA ART ÛÜ(àK}~hijk~#
u 3 ÌâBu BCDEFGH é7ÍYþÎ EFAÏÐ 512 512Ñ 256 Ò§2"Lena""Baboon""Jet" "Pepper"ÓÔâBunu ùvrK@ÍY ÌãA ÕÓ nPSNR,dBr`nCR, bits/pixelr@ÍY# ( !"#$ $"Lena"tÐâBuùoP2é7ÍY ö 1 `ö 2 ùv)5Ç2âBu2éê` Çö 345 ùv# ðþÎ EF¤6Ö×UÈÊ2ef iQVUÍÊ2HIJ)ä¿TÒ Õ Ó st¿¡0ÍË2# ¢EF2&'(ghû }~ØÙý¤ Æ8>5ÇÏ-Ú2HIJZ[#
PSNRdB CRb/p SOFM 29.02 1.75 AMLP 28.40 1.0 PHNN 29.22 1.03 SHNN 31.10 1.75 1 ”Lena”SOFMAMLP PHNN SHNN Vigilance ρ # of subnets PSRN (dB) CRv(b/p) 0.300 57 25.60 0.65 0.325 89 25.88 0.73 0.350 145 26.67 0.82 0.375 225 27.07 0.85 0.400 357 28.10 0.98 0.425 645 28.76 1.18 0.450 1278 30.94 1.56 2 ”Lena”ART-based Vigilance ρ # of subnets PSRN (dB) CRv(b/p) 0.300 113 19.28 0.74 0.325 208 19.88 0.85 0.350 377 20.42 0.99 0.375 644 20.77 1.18 0.400 1066 21.24 1.45 0.425 1631 21.96 1.73 0.450 2435 23.87 2.19 2 ”Baboon”ART-based Vigilance ρ # of subnets PSRN (dB) CRv(b/p) 0.300 79 24.52 0.72 0.325 122 25.47 0.75 0.350 182 26.23 0.83 0.375 296 26.42 0.88 0.400 459 27.72 1.03 0.425 750 28.97 1.24 0.450 1333 31.09 1.58 2 ”Jet”ART-based Vigilance ρ # of subnets PSRN (dB) CRv(b/p) 0.300 76 25.12 0.72 0.325 108 25.68 0.74 0.350 162 25.80 0.82 0.375 247 26.85 0.87 0.400 366 27.99 0.99 0.425 634 29.22 1.18 0.450 1299 30.76 1.57 2 ”Pepper”ART-based 78 IJKL )*#$ $"Lena"tÐâBuùoP2é7ÍY ö 6 ùv#i>EFGIÊ2HIJ[ `0-82HIJ[EFù2 }~Û,I FSRVQ Z[2Ü`Æݯ ý¤sEFéê¢ÞúI FSRVQ Z[# ¡0rßê2Iq> FSRVQ HIJZ[ 2áÇ2HIJ[ 8àÇ2 áÊ2âàQAéêãÒs¿äÊ}~ 2|Ö# EFA §~$tstHIJ Z[2éêQRÊ2TãstÞ$ I FSRVQ Z[)*ÇGµ¶ 2}~ )EF2Z[à Í8u# EFGtu82HIJ`¯à ÄÅ¢¤}~á-HIJ[åS æ}~tQAoEF2Z[ù2}~ PSNR (dB) CR (b/p) FSRVQ 30.65 0.212 FSRVQ 29.41 0.212 6 ”Lena”FSRVQ !" FSRVQ MGN EFTU2HIJZ[,tn6 2HIJ&'(+TU2-HIJ [nç )äEF2(Ô¡Y CݯúH.YCù6EF2 (+TU2Ï-HIJZ[ c þ# OPQR
1. M. A. Abidi, S. Yasuki and P. B. Crilly, Image compression using hybrid neural networks combining the auto-associative multi-layer perceptron and the self-organizing feature map. IEEE Trans. On Consumer Electronics. 40(4): 796-811, November.
2. S. A. Rizvi, N. M. Nasrabadi, Finite-state residual vector quanization using a tree-structured competitive neural network, IEEE Trans. On Circuits and Systems for Video Technology, 7(2): 377-390, 1997.
3. G. W. Cottrell and P. Munro, Principal components analysis of Images via backpropagation. SPICE vol. 10011, Visual Communication and Image Processing, pages:1070-1077, 1988.
4. N. M. Nasrabadi and R. A. Feng. Vector quantization of Images based upon Konhnen self-organizing feature map. IEEE Int. Conf. On Nerual Networks, 1:101-108, 1988.
5. M. Mougeot, R. Azencott, and B. Angeniol. Image compression with backpropagation: Improvement of the visual restoration different cost functions. Neural Networks, 4:467-476, 1991.
6. H. L. Tsai, S. H. S un, and S. J. Lee. Image compression using ART-based neural networks. In proceedings of National Computer Symposium, 1:B163-B168, Taichung, Tainwan, 1997.
