Ly Thuyet Thi

  • View
    36

  • Download
    7

Embed Size (px)

DESCRIPTION

Tham kho

Transcript

L THUYT H C S D LIU A PHNG TIN

Lp: D11CQTT01 NTn: V Th Kim HinM SV: N1121040301. Trnh by c ch nn Huffman.

Xp tt c cc biu tng theo th t xc sut xy ra. Xc nh v tr hai biu tng vi xc sut nh nht. Thay th hai biu tng ny bng mt biu tng mi, c xc sut l tng ca 2 biu tng thnh phn. To 1 nt cha ca 2 biu tng.

Lp li cc bc 1, 2, 3 cho n khi ch c 1 biu tng. Cui cng, chng ta s c mt cy m mi nt l tng xc sut ca tt c cc nt l bn di n. Duyt qua cy t gc n tng l, nh s 0 cho nhnh tri v 1 cho nhnh phi.

2. Phn bit tm kim i tng multimedia theo thuc tnh v theo ni dung.THUC TNHNI DUNG

c imTm kim nh da vo cc c trng trong vn bn i km vi nh. (tn file, li ch thch, cc t kha keyword,).Tm kim nh da trn ni dung ca nh. (Mu sc, hnh dng v kt hp c hai)

u imn gin.Nhanh c kt qu.

Nhc imDa trn nhng c trng mang tnh ch quan v khng c tnh duy nht.Ph thuc vo tng ngn ng khc nhau.

Ph thuc vo tng trng hp a ra cu lnh tm kim cho ph hp.

Truy vn ( chnh xc khng cao.Kh khn trong vic phn tch ni dung nh.Cn khng gian lu tr ln lu tr d liu cho rt trch.

3. Trnh by phng php tm kim i tng nh theo ni dung.M hnh tm kim i tng nh theo ni dung: u tin, nh u vo s qua giai on tin x l rt trch cc tnh cht c trng, sau , cc tnh cht c trng ny s c lu tr cng vi nh trong CSDL. Khi c 1 nh cn truy vn th nh ny cng phi qua giai on tin x l nh trn rt trch cc tnh cht c trng, sau , n s so trng vi cc vector c trng c trong CSDL. Cui cng, n s cho ra 1 tp nh c trng khp cao.

4. Trnh by cch so trng 2 i tng da trn o tng t. Cho v d.Mt o tng t thch hp gia mt vector c trng hnh nh F v vector truy vn Q l ma trn trng s W:W = (F Q)T . A . (F Q)trong A l mt ma trn nxn m c th c s dng xc nh cc o trng s ph hp.

S minh ha: cc im nm cng mt khong cch t im truy vn l tt c u tng t, v d nh, F1 v F2.

Cch xc nh trng s: ty vo tng trng hp c th m ngi ta c th da vo kinh nghim cho gi tr trng s v thu v kt qu chnh xc nht.5. Trnh by cu trc tm kim a chiu theo phng php ng cong lp y khng gian.ng cong lp y khng gian (space filling curves) l ng cong i qua tt c cc im trong 1 khng gian a chiu. Gi s mi chiu c i din bi 1 s chiu rng bit c nh. Phn vng khng gian vi 1 ci li

Nhn ca mi li cha 1 s duy nht c gi l gi tr ng cong

i vi cc im, lu tr s trong mt ch mc mt chiu truyn thng Cc i tng c th c x l thng qua phn tch thnh nhiu

Cc loi ng cong lp y khng gian: Fractals, Hilbert curve, Z Order curve6. Trnh by cu trc tm kim a chiu theo phng php cy D-tree. D-tree l PP tch min. Nu s lng i tng nm trong min vt qu 1 ngng nht nh, th cc tn min s c chia thnh 2 tn min ph.

C 2 kiu chia l theo chiu dc v theo chiu ngang.

Node trong s cha tn min trc khi chia. Node ngoi cha cc tn min sau khi tch t node trong. (VD: khi cha chia ta c min D ( node trong (internal node) s cha D. Sau khi chia D thnh D1,D2 ( node ngoi( external node) s cha D1,D2. Nu chia nh min )

7. Trnh by cu trc tm kim a chiu theo phng php cy Point quad tree. Mi im trong cy s chia 1 vng thnh 4 vng con theo c 2 chiu nang v dc

NW (Northwest)

SW (Southwest)

NE (Northeast)

SE (Southeast)

Mi nt trong cy ngm biu din 1 vng

8. Trnh by cu trc tm kim a chiu theo phng php cy R-tree.

Phn chia khng gian d liu bng cc hnh ch nht ti thiu (MBR Minimum Bounding Rectangles) Cc vng c th chng nhau (overlapped)

D liu lu cc nt l, mi nt l cha nhiu d liu (t chc d liu trong mi nt l l ty chn)

Mi R- tree c bc K: mi nt trong ca cy (tr nt gc) cha nhiu nht K vng v t nht K/2 vng

C 2 loi vng:

Real rectangles (vng nt l)

Group rectangles (vng nt trong)

Tm kim c th phi thc hin theo nhiu nhnh do c s chng cho ca cc MBR

Hnh dng cy ph thuc vo th t d liu thm vo

Hiu qu cho lu tr d liu ln trn a

Cch hiu qu ti thiu s ln truy nhp a (qun l d liu theo vng)9. Trnh by cu trc tm kim a chiu theo phng php cy K-Dtree. L dng m rng ca cy nh phn dnh lu tr d liu im a chiu (k dimension). VD: 2 tree: lu d liu im 2 chiu

Mi im l vector c k phn t

Khng lu d liu vng

mi mc, cc bn ghi s c chia theo gi tr ca 1 chiu nht nh (discriminator)

Cy dng xy dng ph thuc vo th t cc im c a vo

Cy khng cn bng

10. Trnh by tm kim nh da trn ni dung theo 3 cch:

Query by example: l ch mt hnh nh v hy vng rng h thng tr v tt c cc loi bm tng t nh th. c im: tm kim n nh Thng dng trong cc cu truy vn cui cng v khi ngi dng mun tm thm cc kt qu gn ging vi kt qu trc Tr v kt qu hnh nh: cc con bm tha iu kin truy vn. i vi nh rt ging nhau: tr v ht tt c i vi nh hi ging nhau: tr v nh i din Query by feature: l chn vi c trng ca i tng (VD: con bm) v hy vng h thng s tr v tt c cc con bm c nhng c trng VD: c trng ca con bm: mu ch o, hnh dng, m hnh kt cu c im: tm kim th (c chnh xc khng cao) c dng trong truy vn ln u v ngi dng mun m rng s quan st trong khng gian tm kim Tr v cc c trng lin quan n cc iu kin truy vn trc . Kt qu tr v ch l cc con bm i din Nu truy vn theo 1 c trng th kt qu hnh nh tr v l nhng hnh c trng khp > 0 v c xp theo mc trng khp . Cn truy vn theo nhiu c trng th chng ta s hp nht cc chui c trng tng ng Feature vertor: Mc tiu: tm kim hiu qu Vn ch mc trong CBIR (Content-based Image Retrieval) Chiu ca khng gian c trng l rt ln

Cu trc ch mc c th h tr cc o tng t Euclidean and non-Euclidean Gii php:

Gim s chiu: KLT, DCT, DWT. Ch mc tng t: R*-tree, SS-tree, SR-tree.11. Trnh by h thng truy vn d liu nh phi nhiu. Cc k thut truy vn nh phi nhiu.Truy vn nh phi nhiu c chnh xc cao hn. Ngi dng c th xc nh cc rng buc v khng gian v t l ca i tng.H thng truy vn nh phi nhiu:

Giai on chun b: Ly mu nh truy vn theo t l khc nhau. i vi mi t l ly mu, thc hin cc bc:

Xc nh vng chnh (core area) cha s lng ln nht ca cc khi ly mu c lin quan vi nhau v t nhiu nht.

Xc nh vng (hnh ch nht) cn truy vn Tnh ton gi tr ca vng chnh (core area) Giai on tm kim: S dng vng (hnh ch nht) cn truy vn tm cc cm c lin quan trong R* tree

Dng du hiu c trng nh (short signature) loi b subimages khng lin quan Mi subimage s phi qua cc bc test trn v c so snh vi nh phi nhiu ban u bng cch kt hp cc khi ly mu tng ng.

Mi hnh nh c mt subimage thch hp s c ly ra. Tm li, cc bc chnh gm c: Ly mu NFQ theo t l khc nhau Xc nh vng chnh (core area) v tnh ton du hiu c trng ca n Xc nh hnh ch nht truy vn Tm kim cc cm c lin quan (hoc MBRs) trong R* tree

Loi bt cc subimage thng qua du hiu c trng nh (short signature)Cc k thut truy vn nh phi nhiu:

Local Color Histogram (LCH) Mi hnh nh nh u c 1 histogram mu.

Bt k s kt hp ca cc histogram c th c chn so snh vi cc histogram mu tng ng ca nh truy vn. Hn ch:

Vn : Nu xc nh cc vng ln th khng chnh xc, cn xc nh cc vng nh th qu tn km Kh xc nh t l Correlogram (lc tng quan): m t phn phi mu ca cc im nh v s tng quan v khng gian gia cc cp mu. SamMatch: ly mu hnh nh lin quan n cc khong khng gian. i vi mi vng ly mu, ta tnh ton cc mu trung bnh ca cc im nh trong vng. Cc gi tr trung bnh ny to ra cc vector c trng ca hnh nh.Kt lun:

Gim nhiu l iu cn thit t c truy xut hnh nh ng tin cy hn

SamMatch h tr NFQs rt hiu qu: nhanh hn so vi LCH gp 2 ln, nhanh hn correlogram gp 1 ln Cc li ch khc ca SamMatch bao gm:

So snh i tng cc t l khc nhau

Pht hin s tnh tin ca cc vng ph hp

Gii quyt cc rng buc v khng gian v t l SamMatch s dng t hn 1/16 khng gian lu tr cn thit so vi LCH v correlogram

12. Cho bit cc k thut pht hin shot. nh gi cc k thut ny da trn 2 o (recall, precision):

Pht hin Shot da vo:

c trng ni dung frames v nhm chng li: phc tp + thi gian ln c trng ca camera: gc quay, s chuyn ng: kh

Hiu ng ca s chuyn i gia cc cnh quay: thng c s dng (nh Shot Boundary Detection)

Phng php Shot Boundary Detection:

Color Histogram:

Lu tr cc t l phn trm mu ca frame Cc kt qu c so snh vi cc frame lin k

Tnh ton gi tr chnh lch

Nu gi tr chnh lch vt qua 1 ci ngng xc nh (threshold) th chnh l s thay i shot

Edge Detection: Chuyn cc frame thnh nh mc xm

p dng thut ton Edge Detection cho cc nh Tnh ton gi tr chnh lch gia 2 frame lin k Nu gi tr chnh lch vt qua 1 ci ngng xc nh (threshold) th chnh l s thay i shot Macroblock (compressed domain): Dng trong cc video k thut s nn MPEG.

Frame chia thnh cc vng c nh gi l macroblocks. C 3 loi macroblocks:

I: ch nhng vng c nh khng thay i g P: ch nhng vng b thay i so vi frame trc B: ch nhng vng b thay i so vi frame trc v c frame sau Pht hin shot thay i da vo con s c th ca cc loi macroblock s xy ra. Spatio-Temporal Slice Model: pht hin shot khi c s vi phm ca mu kt cu mch lc (color-texture coherency ) trong cc lt ct khng gian - thi gian rt trch t mt chui video.

nh gi cc k thut da trn 2 o:

Hn ch ca Shot Boundary Detection: Kh khn xc nh cc thng s u vo.

chnh xc thay i t 20% n 80%

Tnh ton l tn km.( Thm k thut pht hin shot: Background Tracking (BGT), Non-Linear Approach to Shot Boundary Detection

u im ca Background Tracking (BGT):

t bin i vi ngng

ng tin cy hn

t tnh ton

Non-Linear Approach to Shot Boundary Detection ( Gim ng k s ln so snh Regular Skip: so snh u n cc frame theo 1 khong cch d nht nh.

VD: so snh cc frame chn vi nhau v cc frame l vi nhau.

Adaptive Skip: so snh cc frame theo 1 khong cch d thay i 1 cch ti u nht. Binary Skip: so snh frame u & cui, sau chia chui frame ra lm i v tip tc thc hin nh vy vi tng chui frame con, cho n khi tha iu kin cui cng th dng.13. Nu phng php xc nh key frame trong 1 shot.Xc nh key frame thng l nh gi ch quan. Key frame rt ra t mi shot i din cho ni dung ni bt ca cc shot.Phng php xc nh key frame:

Cc frame trong 1 shot c phn on t video bi thut ton Shot Boundary Detection v c phn ra thnh N cm theo gi tr khng lin tc ca cc frame. Cc cm ng vin s c chn ra t t