NHN DNG S QUA CAMERA GVHD:NH QU TH

MC LCLi ni u..2

Phn mt:L THUYT X L NH...3

Chng 1:Gii thiu h thng x l nh3Chng 2:Thu nhn nh.4

Chng 3:Phn tch nh.........6Chng 4:Nhn dng nh .....21Phn hai:THC NGHIM24Kt lun.30Ti liu tham kho.30

LI M U

Trong my thp k gn y x l nh c nghin cu mnh m v c nhiu ng dng thc t. Nh trong y hc x l nh c dng pht hin v nhn dng khi u, ci thin nh X quang, nhn dng ng bin mch mu t nh chp bng tia X. Trong truyn thng v trong nghin cu v tr x l nh c dng phn tch nh ca nhng hnh tinh , thin h thu c t tu v tr hay knh thin vn. c bit trong Robot ngy nay khng th thiu yu t x l nh , nh x l nh robot c th pht hin v nhn dng i tng ngoi mi trng . T gii quyt cc bi ton trnh vt cn ,tm ng, nhn dng i tng Qu trnh x l nh c chia lm 3 bc chnh sau y: thu nhn nh , phn tch nh v quyt nh.Ni dung ca n :Nhn dng cc ch s qua camera kt ni vi my tnh. C th tm tt ni dung n qua lc c bn sau y:

Gii hn thit k n :Do x l nh l mt lnh vc mi v kh, khng c nhiu ti liu trong nc tham kho nn chng em thit k n ny ch gm 1 webcam kt ni my tnh thng qua toolbox x l nh ca Matlab thc hin.Mc ch ca n:Tm hiu v qu trnh x l nh t qu trnh thu thp nh , lc nhiu , phn on nh , tch bin , n qu trnh nhn dng nh . Sn phm ca n trc ht c ng dng vo mc ch nghin cu v ging dy ti trng , nu c nghin cu v i su c th c ng dng vo thc t nh h thng nhn dng s chng minh th nhn dn , bin s xe PHN MT:L THUYT X L NH

CHNG MT:GII THIU H THNG X L NHMt h thng x l nh c bn bao gm ba bc chnh : thu thp nh , phn tch v x l nh , x l kt qu , c th tm tt bng s c bn sau : SHAPE \* MERGEFORMAT

1.1 Thu nhn nh:Vic thu nhn nh c th thng qua camera . Camera c th l loi tng t hoc l loi s . Nu nh thu c l loi tng t th n phi c ly mu v lng t ha trc khi phn tch , x l .1.2 Phn tch nh:

giai on ny nh c x l qua nhiu cng on nh nh lc nhiu , ci thin nh , khi phc nh lm ni mt s c tnh chnh ca nh hay lm cho nh gn vi nh gc , tip theo l pht hin cc c tnh bin cnh gc ,phn vng.

1.3 Nhn dng nh:

Nhn dng l qu trnh xc nh nh . Qu trnh ny thng c xc nh bng vic so snh vi mu c t trc . Theo l thuyt v nhn dng nh c hai loi nhn dng sau:-Nhn dng theo tham s.

-Nhn dng theo cu trc.

Mt s i tng nhn dng kh ph bin hin nay c p dng trong khoa hc v cng ngh l : nhn dng k t , nhn dng vn tay , nhn dng vn bn , nhn dng mt ngi

CHNG HAI: THU NHN NH

2.1 Thit b thu nhn nh:Chc nng ca thit b ny l s ha mt bng tn s c bn ca tn hiu ca tn hiu vo t mt camera hay mt u my thu tn hiu nh .nh thu c c th l mt nh en trng hoc nh mu , nh thu c thng l loi tng t cng c loi camera c s ha l loi photodiot to cng sng ti mi im nh.

Cht lng nh thu c ph thuc vo cht lng thit b thu v mi trng (nh sng ,phong cnh).

2.2 H ta mu:

T chc quc t v chun ha CIE a ra mt s cc chun biu din mu . Cc h ny c cc chun ring y cp h ta mu CIE-RGB ( h ta dng 3 mu c bn l red , green , blue ) mi pixel Px c vit: Px =

Ngi ta dng h ta 3 mu R-G-B (tng ng vi 3 trc ta x,y,z) biu din mu nh sau: Blue(l)

(0,0,1) l (0,0,1) tm xanh

tm (1,0,1) (1,1,1) vng m

(0,0,0) en (0,1,0) lc Green (lc)

(1,0,0) (1,1,0) vng

Red()

Trong cch biu din trn +lc + l =1 , cng thc ny gi l cng thc Maxell , trong hnh v trn tam gic to bi ba ng t on l tam gic Maxell.2.3 Ly mu v lng t ha:

2.3.1 Ly mu:

Ly mu l mt qu trnh trong nh c to nn bi mt vng lin tc c chuyn thnh cc gi tr ri rc nguyn . Qu trnh ny gm 2 bc l:

-Khong ly mu.

-Cch th hin dng ly mu.

Khong ly mu cng nh th nh thu c cng ging vi nh nguyn gc .

Dng ly mu l cch bi tr cc im ly mu trong khng gian 2 chiu . Mt s dng ly mu im nh c cho l dng tam gic , ch nht , lc gic. 2.3.2 Lng t ha:

Lng t ha l nh x t cc s thc m t gi tr ly mu thnh di cc s thc , ni cch khc l qu trnh s ha bin .

2.4 Biu din nh:

nh thng c biu din bng mt s phng php sau y:

Biu din m lot di (Run length code) .

Biu din m xch ( Chain code ).

Biu din m t phn (Quad tree code ).

2.4.1 Biu din m lot di:Thng dng biu din nh nh phn , mt vng nh R c th biu din n gin nh mt ma trn nh phn :

u(m,n) = 1 nu (m,n) ( R

0 nu khc

2.4.2 Biu din m xch:Thng biu din bin ca nh . Thay v lu tr ton b ngi ta lu tr cc dy im nh A, B,.M . 2.4.3 Biu din m t phn:

Theo phng php ny mt vng nh coi nh bao kn mt hnh ch nht . Vng ny c chia lm 4 vng con nu mt vng con ny gm ton im en (1) hoc im trng (0) th khng cn chia tip ngc li th li chia vng ra lm 4 vng nh , c tip tc chia n khi no vng ng nht ch gm im en hoc im trng th thi . Qu trnh to thnh mt cy chia theo 4 phn gi l cy t din .CHNG 3 : PHN TCH NH

3.1 Khi nim pixel v pixel ln cn :Pixel l phn t nh nht ca nh cu to nn nh . Mi pixel c ta p(x,y) v c mu xc nh. SHAPE \* MERGEFORMAT

Hnh minh ha :Pixel P v cc ln cn ca P

Mt pixel P(x,y) c 4 pixel ln cn theo chiu dc v chiu ngang l P2(x-1,y) ; P4(x,y-1) ; P5(x,y+1) ; P7(x+1,y); chng k hiu l N4(p).Ngoi 4 pixel ny cn 4 pixel ln cn cho gc l P1(x-1,y-1) ; P3(x-1,y+1) ; P6(x+1,y-1) ; P8(x+1,y+1) ; chng k hiu l ND(p)

Tp 8 pixel trn to thnh ln cn ca P(x,y) k hiu l N8(p).

3.2 Mt s cng c tr gip x l nh: 3.2.1 Tch chp:

Ton t chp c nh ngha nh sau:

+ trng lin tc g(x,y) = h(x,y) ( f(x,y) =

+ trng ri rc y(m,n) = h(m,n) ( x(m,n) = 3.2.2 K thut lc s: Cht lng hnh nh km do nhiu nguyn nhn gy ra nh do nhiu in t ca my thu hoc do cht lng b s ha km . Phn ny tm hiu v cc b lc c dng trong x l nh khc phc s nhiu ca nh thu c.C nhiu b lc c thit k x l nh nhiu nhng do gii hn ca n mn hc nn phn ny khng c gii thiu y.

3.3 Cc php bin i nh:

- S biu din ton hc cho mt nh thng l mt hm vi hai bin khng gian f(x,y) . Gi tr ca hm ti mt v tr ( x,y ) i din cho cng ca nh ti v tr . Thut ng bin i ( transform ) nhm ni n mt s biu din ton hc tng t ca nh .

3.3.1 Php bin i Fourier ( phn ny dch t matlab )- Php bin i Fourier l mt biu din ca nh nh l tng ca cc hm m phc ca bin , tn s v pha bin i . Bin i Fourier chim mt vai tr quan trng trong cc ng dng x l nh bao gm : lm giu nh ( hay ci thin cht lng nh enhancement ) , phn tch , phc hi v nn nh .

a - nh ngha php bin i Fourier

- Nu f(m,n) l mt hm vi hai bin khng gian c lp m v n , th bin i Fourier hai chiu ca hm f(m,n ) c nh ngha bi quan h :

Bin v l cc bin tn s . Hm c gi l biu din trong min tn s ca hm f(m,n) . N l mt hm phc tun hon vi chu k . Do tnh tun hon , nn v thng c chn trong khong - n . Ch rng F(0,0) l tng ca tt c cc gi tr ca f(m,n) .V l do ny F(0,0) thng c gi l thnh phn khng i hoc thnh phn mt chiu DC ca bin i Fourier .

- Bin i Fourier ngc c cho bi cng thc :

Ni chung , phng trnh ny c ngha rng f(m,n) c th c i din nh l tng v hn ca cc hm m phc vi cc tn s khc nhau . Bin v pha ca thnh phn tn s c lu trong F

Biu din trc quan bin i Fourier

- minh ho , ta hy xem hm f(m,n) nhn gi tr bng 1 trong khong hnh ch nht ( xem hnh ) v bng 0 ti mi im khc . n gin s f(m,n) c coi nh mt hm lin tc mc d m, n l cc bin ri rc.

- Hnh biu din sau y biu din bin ca bin i Fourier ca hm ch nht nh trn .

- Gi tr nh tm ca th l F(0,0) , l tng ca tt c cc gi tr ca hm f(m,n) . th cng ch ra rng c nhiu nng lng hn vng tn s ngang so vi tn s ng . iu ny phn nh s thc rng cc vng giao ngang ca hm f(m,n) ( horizontal cross sections ) l cc xung hp trong khi cc vng giao ng ca hm ny l cc xung rng . Xung hp mang nhiu ni dung tn s hn xung hp .

Mt cch khc biu din trc quan bin i Fourier l hin th hm log nh mt nh :

- S dng hm lgarit gip cho vic nhn r c tnh ca bin i Fourier cc vng m gi tr gn bng 0 . minh ho , hy xem biu din trc quan cc bin i Fourier sau :

b- Bin i Fourier ri rc

-Bin i Fourier trn my tnh l bin i Fourier ri rc ( DFT ) . C hai l do chnh s dng loi bin i Fourier ny :

+ Hm vo v ra ca bin i Fourier l cc hm ri rc , iu ny thch hp cho cc thao tc bin i trn my tnh .

+ C mt gii thut nhanh cho vic tnh ton DFT c gi l bin i Fourier nhanh ( FFT )

- DFT thng c nh ngha cho cc hm ri rc f(m,n) khc 0 v m , n ln lt nhn cc gi tr ,

- Matlab s dng cc hm fft , fft2 v fftn s dng thut ton bin i Fourier nhanh cho vic tnh ton DFT mt chiu , hai chiu v N chiu tng ng . Cc hm ifft, ifft2 v ifftn tnh ton DFT ngc .

Quan h vi bin i Fourier

- Ta c quan h biu din bng cng thc sau :

V d :

1. To ma trn f tng t nh hm f(m,n) ch nht xt trc y . Hm f(m,n) bng 1 trong vng ch nht v bng 0 trong vng khc . S dng mt nh nh phn thay th cho f(m,n)

f = zeros(30,30);

f(5:24,13:17) = 1;

imshow(f,'notruesize')

2. Tnh ton v biu din trc quan bin i Fourier bi nhng lnh sau y :

F = fft2(f);

F2 = log(abs(F));

imshow(F2,[-1 5],'notruesize'); colormap(jet); colorbar

3 . thu c kt qu tt hn , ta s dng cc lnh sau y :

F = fft2(f,256,256);

imshow(log(abs(F)),[-1 5]); colormap(jet); colorbar

c- Cc ng dng s dng php bin i Fourier

p ng tn s ca b lc tuyn tnh

- Bin i Fourier ca p ng xung ca b lc tuyn tnh cho php nhn c p ng tn s ca b lc . Hm freqz2 tnh ton v hin th p ng tn s ca mt b lc . p ng tn s ca b lc Gauss ch ra rng b lc ny l mt b lc thng thp .

h = fspecial('gaussian');

freqz2(h)

Nhn chp nhanh

- Mt tnh cht quan trng ca bin i Fourier l nhn hai bin i Fourier tng ng vi nhn chp (convolution ) ca hai hm kt hp trong khng gian . Tnh cht ny kt hp vi bin i Fourier nhanh lm c s cho gii thut nhn chp nhanh .

Ch : FFT da trn phng php nhn chp thng c s dng vi u vo ln . Vi cc u vo nh , nhn chung s dng hm imfiler s nhanh hn .

- minh ho , v d sau thc hin nhn chp ca A v B trong A l mt ma trn MxN v B l ma trn PxQ

1. To ra hai ma trn

A = magic(3);

B = ones(3);

2. Thm cc gi tr 0 vo A v B chng c chiu t nht l (M+P-1)x(N+Q-1) ( ch rng : hm fft2 s nhanh hn nu cc kch thc ca A v B l lu tha ca 2 ) .V d sau chn thm cc phn t 0 hai ma trn c kch thc 8x8

A(8,8)=0;

B(8,8)=0;

3. Tnh bin i Fourier hai chiu ca A v B s dng hm fft2

4. Nhn hai bin i Fourier li vi nhau

5. Tnh bin i Fourier ngc ca kt qu trn bng hm ifft2

C = ifft2(fft2(A).*fft2(B));

6. Trch ra thnh phn khc 0 ca kt qu v loi b phn o do sai s lm trn

C = C(1:5,1:5);

C = real(C)

C =

8.0000 9.0000 15.0000 7.0000 6.0000

11.0000 17.0000 30.0000 19.0000 13.0000

15.0000 30.0000 45.0000 30.0000 15.0000

7.0000 21.0000 30.0000 23.0000 9.0000

4.0000 13.0000 15.0000 11.0000 2.000

Pht hin chi tit ca nh

- Bin i Fourier c th c s dng thc thi tng quan c lin h gn vi nhn chp ( convolution ) . Tng quan c th c s dng pht hin chi tit trong mt nh , trong trng hp ny , tng quan thng c gi l hp mu ( template matching )

- V d sau minh ho vic s dng tng quan pht hin s c mt ca k t a trong mt nh c ch :

1. c vo nh mu :

bw=imread(text.png);

2. To mt mu ( template ) lm hp bng cch trch k t a t nh :

a=bw(32:45,88:98);

Ta c th to nh mu bng cch s dng phin bn tng tc ca hm imcrop , s dng hm pixval tnh to ca mt chi tit trong nh . Hnh sau y ch ra nh gc v nh mu :

3. Tnh ton tng quan ca nh mu a vi nh gc bw bng cch quay nh mu 180 v sau dng bin i Fourier nhanh trn c s php nhn chp nh cp trc y ( Nhn chp s tng ng vi tng quan nu ta quay nhn chp 180 ) . hp vi nh mu , dng hm fft2 v ifft2 .

C = real(ifft2(fft2(bw) .* fft2(rot90(a,2),256,256)));

nh sau y minh ho kt qu ca tng quan . m sng trong nh tng ng vi s xut hin ca k t :

figure, imshow(C,[]) % Scale image to appropriate display range.

4. quan st v tr ca mu trong nh , tm gi tr pixel ln nht sau nh ngha mt gi tr ngng nh hn gi tr ny . V tr ca nh c ch ra bng mt im trng trong nh tng quan mu ( xc nh chng d hn , nh ngng c m rng kch thc im )

max(C(:))

ans =

68.0000

thresh = 60; % Use a threshold that's a little less than max.

figure, imshow(C > thresh)% Display showing pixels over threshold.

3.3.2 Bin i c-sin ri rc

- Bin i c-sin ri rc ( DCT ) biu din mt nh nh l tng ca cc hm sin vi bin v tn s bin i . Hm dct2 tnh DCT hai chiu ca mt nh . DCT c tnh cht m vi cc nh in hnh , hu ht cc thng tin c ngha v nh c tp trung vo cc h s ca DCT . V l do ny , DCT thng c s dng trong cc ng dng nn nh . Chng hn ,DCT l trung tm ca gii thut nn nh theo chun quc t thng c bit vi tn JPEG ( tn ny do nhm pht trin t ra : Joint Photographic Experts Group )

- DCT hai chiu ca ma trn A c kch thc MxN c nh ngha nh sau :

- Cc gi tr Bpq c gi l h s DCT ca A . DCT c th bin i ngc c v bin i ngc ca n cho bi cng thc :

- Biu thc DCT ngc c th c xem xt khi coi rng mi ma trn A kch thc MxN nh l tng ca MN hm c dng :

- Nhng hm ny c gi l nhng hm c s ca DCT . H s DCT Bpq c th c xem nh trng s cho mi hm c s . Vi cc ma trn 8x8 , 64 hm c s c minh ho bi nh sau :

Ma trn bin i DCT

- Toolbox x l nh s dng 2 cch tnh DCT . Cch th nht l dng hm dct2 . Hm dct2 s dng gii thut da trn FFT tng tc tnh ton vi cc nh c kch thc ln . Cch th hai l s dng ma trn bin i DCT . Ma trn ny c tr v t hm dctmtx v c s dng hiu qu hn vi cc nh c kch thc nh nh 8x8 , 16x16 . Ma trn bin i MxM T c cho bi :

Vi ma trn A c kch thc MxM , T*A l mt ma trn MxM m cc ct ca n l DCT mt chiu ca cc ct trong A . DCT hai chiu ca A c th thu c t biu thc : B=T*A*T . Do T l ma trn trc giao nn ma trn nghch o ca n cng l ma trn chuyn v nn DCT hai chiu ca B l T*B*T

DCT v nn nh

- Trong gii thut nn nh JPEG , nh vo c chia lm cc khi c kch thc 8x8 hoc 16x16 v DCT hai chiu c tnh cho mi khi . H s DCT sau c lng t ho , m ho v truyn ti . B nhn JPEG ( hoc chng c file JPEG ) s gii m h s c lng t ca DCT , tnh DCT ngc cho mi khi v sau t cc khi li vi nhau trong mt nh duy nht . Vi cc nh in hnh , nhiu h s DCT c gi tr gn 0 , nhng h s ny c th b b qua m khng nh hng nhiu n cht lng ca nh nn

- V d sau tnh DCT hai chiu ca cc khi 8x8 trong mt nh , b qua ( t bng 0 ) tt c ngoi tr 10 trong s 64 h s DCT ca mi khi v sau xy dng li nh bng DCT hai chiu ngc ca mi khi . Phng php ma trn chuyn i DCT c s dng :

I = imread('cameraman.tif');

I = im2double(I);

T = dctmtx(8);

B = blkproc(I,[8 8],'P1*x*P2',T,T');

mask = [1 1 1 1 0 0 0 0

1 1 1 0 0 0 0 0

1 1 0 0 0 0 0 0

1 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0];

B2 = blkproc(B,[8 8],'P1.*x',mask);

I2 = blkproc(B2,[8 8],'P1*x*P2',T',T);

imshow(I), figure, imshow(I2)

- Mc d cht lng ca nh nn b suy gim , n vn r nt thm ch gn 85% h s DCT b b qua 3.4 X l v nng cao cht lng nh:Nng cao cht lng nh l mt bc quan trng , to tin cho x l nh . Mc ch chnh l nhm lm ni bt mt s c tnh chnh ca nh nh thay i tng phn , lc nhiu , ni bin , khuych i nh ,Khi phc nh l nhm lm cho nh gn vi nh gc trc khi b bin dng do nhiu nguyn nhn khc nhau.

3.4.1 Cc k thut tng cng nh:Tng cng nh l lm ni bt cc c trng chn , lm sao c th pht hin sm hn , bao gm cc bc : iu khin mc xm , dn tng phn , gim nhiu , lm trn nh , ni bin , 3.4.1.1 Ci thin nh dng ton t im:-Tng tng phn

-Tch nhiu v phn ngng

-Bin i m bn

-Ct theo mc

-Trch chn bt

-Tr nh

-Nn di sng

-M hnh ha v bin i lc xm

3.4.1.2 Ci thin nh dng ton t khng gian:-Lm trn nh bng lc tuyn tnh

-Lm trn nhiu bng lc phi tuyn

-Mt n g sai phn v lm nhm

-Lc thng thp lc thng cao v lc di thng

-Khuch i v ni suy nh

-Ci thin nh nh phn-Co , dn nh

3.4.2 Khi phc nh:3.5 Pht hin bin: 3.5.1 Khi nim:-im bin : Mt im nh c coi l im bin nu c s thay i nhanh hoc t ngt v mc xm.- ng bin : L tp hp lin tip ca cc im bin to thnh mt ng bao 3.5.2 Cc phng php pht hin bin nh:

- Phng php pht hin trc tip : da ch yu vo s bin thin sng ca im nh lm ni bin bng cch ly o hm

+ Ly o hm bc nht ta c phng php Gradient

+ Ly o hm bc hai c phng php Laplace

C 2 phng php ny gi chung l phng php d bin cc b

- Phng php pht hin gin tip : Bng cch no y chng ta thu c cc vng nh khc nhau th ng phn cch cc vng nh chnh l bin . Phng php ny kh thc hin nhng cho kt qu tt khi s bin thin mc xm nh .

3.5.2.1 Phng php pht hin bin cc b: a. Phng php Gradient: b. Phng php Laplace:CHNG BN : NHN DNG NH4.1 Gii thiu:Nhn dng l giai on cui ca h thng x l nh . Nhn dng da trn l thuyt v nhn dng , l thuyt ny v l thuyt v nhn dng nh ni ring cp n 3 phng php nhn dng chnh:

Nhn dng da vo phn hoch khng gian

Nhn dng da vo cu trc

Nhn dng da vo mng nron

Khi nim v nhn dng: Nhn dng l qu trnh phn loi i tng c biu din theo mt m hnh no v gn cho chng mt tn da theo quy lut v mu chunBn cht ca qu trnh nhn dng gm 3 giai on chnh:

La chn m hnh biu din i tng Chn ra lut quyt nh ( phng php nhn dng ) v suy din

Hc nhn dng

4.2 Mng nron:Mng nron nhn to c thit k da trn m hnh mng nron thn kinh vi cc m hnh nron nhn to l m phng nron thn kinh.

M hnh ca mt mng nron nhn to gm 3 phn chnh : b tng kt u vo , b hc tuyn tnh v b phi tuyn .

B tng lin kt u vo: L b tng hp cc lin kt ti u vo ca mt phn t nron, m t nh sau:

mV(t)= wk xk(t)- K=1

x1

x2

xm

Hnh 1.2. M hnh nron nhn to.

Trong :

v(t): Tng tt c cc u vo m t ton b th nng tc ng thn nron .

xk(t): Cc u vo ngoi; vi k l ch s chy, k=1,2m.

m: S lng u vo m t tn hiu vo t cc u nhy thn kinh hoc cc nron khc.

wk: Trng lin kt ngoi gia cc u vo k ti nron hin ti.

y(t): Tn hiu u ra nron .

: Ngng (l hng s), xc nh ngng kch thch hay c ch.

B ng hc tuyn tnh: u vo ca phn ng hc tuyn tnh l v(t), u ra ca n l u(t) gi l u ra tng t. Hm truyn tng ng ca phn ng hc tuyn tnh c th m t di dng:

U(s) = H(s)V(s) B phi tuyn: Phn ny s dng hm quan h phi tuyn f(.) cho u ra y, chn tnh hiu u ra. Cc hm u ra thng thng l cc hm c gii hn ngng ph hp vi c im u

Phn loi mng nron:C nhiu cch phn loi mng nron nhn to, ty theo tiu ch ra:- Da vo s lng lp c trong mng, ta phn thnh:+ Mng mt lp+ Mng nhiu lp- Da vo ng truyn tn hiu trong mng, ta phn thnh:+ Mng truyn thng+ Mng phn hi+ Mng t t chc

Khi qut lut hc trong mng nronKhi nim hc trong mng nron c hiu theo hai ngha: Hc v cu trc v hc v tham s.Hc tham s (Parameter Learning)Mc tiu ca vic hc tham s l thay i, cp nht cc trng lin kt. Hu ht cc lut hc tn ti thuc kiu hc tham s. Thng thng, lut hc tham s c chia thnh ba dng chnh, l: Hc gim st, hc khng gim st v hc cng c.Hc c thy (Hc gim st: Supervised Learning): Vi kiu hc ny, ti mi thi im c u vo mng nron th u ra mong mun ca h s c cho sn. C th c th ha nh sau:Mng c cung cp mt tp cc mu (x(1),d(1)), (x(2),d(2)), (x(n),d(n)) l cc cp u vo u ra mong mun. Khi mt u vo mng x(k) c a vo, u ra mong mun d(k) cng c a vo mng. Sai khc gia gi tr u ra thc s y(k) v u ra mong mun d(k) s l c s to tn hiu li mng sa li trng sao cho u ra thc s gn vi u ra mong mun hn.Hc khng c thy (Hc khng gim st: Unsupervised Learning): Vi kiu hc ny, khng c bt k mt thng tin phn hi t mi trng no. Mng phi t tm ra cc mu, c tnh, tnh quy tc, s tng quan trong d liu u vo v tp hp li to u ra. Khi t tm ra cc c im ny, mng tri qua cc thay i v tham s ca n. Qu trnh ny c gi l t t chc.Hc cng c (reinforcement learning): i khi trong mt s trng hp, ta khng bit ht c cc gi tr u ra chnh xc c th hc gim st m ch bit c mt s thng tin nh gi. Khi lut hc da trn nhng thng tin nh gi ny gi l lut hc cng c v thng tin phn hi c gi l thng tin cng c. Lut hc ny ch khc lut hc gim st ch, thng tin phn hi t mi trng ch mang tnh cht nh gi ch khng mang tnh cht dy, c ngha l ch nh gi u ra thc s l tt hay xu m khng a ra c u ra mong mun l g. Tn hiu cng c c mng s dng vi hi vng l u ra thc s c nh gi phn hi tt hn trong ln hc tip theo.Hc cu trc (Structure Learning)Mc tiu ca hc cu trc l thay i s nron, kiu lin kt lm cho cu trc mng thay i. i vi hc tham s, chng ta gi s cu trc mng c sau a ra cc thut hc thay i cc trng lin kt nhm lm cho u ra thc s gn vi u ra mong mun. Cn vic hc mc cu trc c th s dng cc k thut lin quan n thut ton gen (Genetic Algorithm) v lp trnh tin ha (Evolutionary Programming). Cc cch tm kim trong thut ton gen v lp trnh tin ha kh tiu tn thi gian ngay c i vi mng c kch thc trung bnh. Do , cn c th s dng cc k thut sa i hay xy dng mng dn dn t mt cu trc ban u. K thut thng c s dng trong trng hp ny l k thut xn bt mng nron, pht trin mng nron v kt hp c hai: Xn bt v pht trin mng nron.PHN HAI : THC NGHIM

1 GII THIU V MATLAB: Matlab khng ch l phn mm m phng h thng m n cn l cng c tnh ton , phn mm lp trnh mnh . N cung cp nhiu toolbox gii quyt cc bi ton ha , iu khin , m phng

Image processing toolbox vi th vin ln cung cp nhiu hm , cc v d demo phc v x l nh

Ngoi ra thu nhn v nhn dng th kt hp 2 toolbox l image acquisition toolbox v neural network toolbox.

2 CHNG TRINH MINH HA: S qu trnh nhn dng s: SHAPE \* MERGEFORMAT

Trong cc bc trn th qu trnh chuyn nh sang nh nh phn quyt nh chnh xc ca qu trnh nhn dng nh . Do nhiu ca mi trng (nh sng) v nhiu ca thit b thu nn sau khi nh phn ha nh thng thng th qu trnh nhn dng t chnh xc khng cao. V vy tng chnh xc khi nhn dng trc khi nh phn ha nh cn x l mc xm tng r nt ca nh nh phn.Sau y l chng trnh minh ha:function varargout = xulyanh(varargin)

% XULYANH M-file for xulyanh.fig

% XULYANH, by itself, creates a new XULYANH or raises the existing

% singleton*.

%

% H = XULYANH returns the handle to a new XULYANH or the handle to

% the existing singleton*.

%

% XULYANH('CALLBACK',hObject,eventData,handles,...) calls the local

% function named CALLBACK in XULYANH.M with the given input arguments.

%

% XULYANH('Property','Value',...) creates a new XULYANH or raises the

% existing singleton*. Starting from the left, property value pairs are

% applied to the GUI before xulyanh_OpeningFunction gets called. An

% unrecognized property name or invalid value makes property application

% stop. All inputs are passed to xulyanh_OpeningFcn via varargin.

%

% *See GUI Options on GUIDE's Tools menu. Choose "GUI allows only one

% instance to run (singleton)".

%

% See also: GUIDE, GUIDATA, GUIHANDLES

% Copyright 2002-2003 The MathWorks, Inc.

% Edit the above text to modify the response to help xulyanh

% Last Modified by GUIDE v2.5 21-Apr-2009 15:44:39

% Begin initialization code - DO NOT EDIT

gui_Singleton = 1;

gui_State = struct('gui_Name', mfilename, ...

'gui_Singleton', gui_Singleton, ...

'gui_OpeningFcn', @xulyanh_OpeningFcn, ...

'gui_OutputFcn', @xulyanh_OutputFcn, ...

'gui_LayoutFcn', [] , ...

'gui_Callback', []);

if nargin && ischar(varargin{1})

gui_State.gui_Callback = str2func(varargin{1});

end

if nargout

[varargout{1:nargout}] = gui_mainfcn(gui_State, varargin{:});

else

gui_mainfcn(gui_State, varargin{:});

end

% End initialization code - DO NOT EDIT

% --- Executes just before xulyanh is made visible.

function xulyanh_OpeningFcn(hObject, eventdata, handles, varargin)

% This function has no output args, see OutputFcn.

% hObject handle to figure

% eventdata reserved - to be defined in a future version of MATLAB

% handles structure with handles and user data (see GUIDATA)

% varargin command line arguments to xulyanh (see VARARGIN)

load mydata;

assignin('base','net',net); %gan gia tri o mydatahandles.output = hObject;

guidata(hObject, handles);

%*****************Phan nay thiet ke giao dien cua chuong trinh************

set(handles.edit3,'string',datestr(clock),'FontSize',16,'FontWeight','bold','ForegroundColor',[1 0 0]);

set(handles.edit1,'string','net');

set(handles.edit2,'string','');

set(handles.text1,'string','So nhan dang duoc la:','foregroundcolor',[1 0 0],...

'backgroundcolor',[0 1 0.5],'fontsize',16,'fontweight','bold');

set(handles.pushbutton1,'string','START','Fontsize',16,'fontweight','bold',...

'foregroundcolor',[1 0 0]);

set(handles.pushbutton2,'string','EXIT','fontsize',16,'fontweight','bold',...

'foregroundcolor',[1 0 0]);

set(handles.uipanel1,'title','Control','fontsize',12,'fontweight','bold');

set(handles.edit2,'fontsize',80,'fontweight','bold','foregroundcolor',[1 0 0]);

% --- Outputs from this function are returned to the command line.

function varargout = xulyanh_OutputFcn(hObject, eventdata, handles)

% varargout cell array for returning output args (see VARARGOUT);

% hObject handle to figure

% eventdata reserved - to be defined in a future version of MATLAB

% handles structure with handles and user data (see GUIDATA)

% Get default command line output from handles structure

varargout{1} = handles.output

% --- Executes on button press in pushbutton1.

function pushbutton1_Callback(hObject, eventdata, handles) %chuong trinh nut start

% hObject handle to pushbutton1 (see GCBO)

% eventdata reserved - to be defined in a future version of MATLAB

% handles structure with handles and user data (see GUIDATA)

set(handles.edit3,'string',datestr(clock),'FontSize',16,'FontWeight','bold','ForegroundColor',[1 0 0]);

%************************THU NHAN ANH*****************************

vid=videoinput('winvideo',1); %tao webcam

set(vid, 'FramesPerTrigger', 8); %so frame thu nhan la 8 giam nhieu

preview(vid);

pause(0.5);

start(vid); %thu anh

wait(vid); %doi thu anh xong

number=vid.FramesAvailable;

[frames,time]=getdata(vid,number);%luu anh vao bien frames

stop(vid);

delete(vid);

clear vid; %xoa webcam

disp(time);

imwrite(frames(:,:,:,6),'data_image.jpg','jpg');%luu anh

so=imread('data_image.jpg'); %doc anh

axes(handles.axes1);

imshow(so),title('original image');%hien thi anh goc tren axes1

%*****************************XU LY ANH*********************************

level=graythresh(so);%lay nguong

%imview(so);

disp(level);

%cai thien muc xam tang do ro cua anh nhi phan

for i=20:200

for j=50:270

if ((so(i,j,1)