PHlTONG PHAP

(1)

MOT S6 PHlTONG PHAP NANG CAO DO PHAN GIAI ANH V I £ N T H A M B A N G PHlTOOVG PHAP SIEU PHAN GIAI

TS. L§ QU6C HUNG, KS. TRLrONG TH! TUY^T Cue Viin tham quoc gia

Nang cao do phan giai khong gian cua anh luon la mdt trong nhung muc tieu nghien cOv h^ng diu cua cac nha khoa hoc trong ITnh vwc viin tham. Bien phap nang cao do phan giai khong gian bing phwang phap nang cip, nghien cwu chi tao cac dau thu, bd cam mai tien tiin cho phep thu anh vai do phan giai khong gian cao co chi phi vd cung Ian. Ben canh do, khi nang cao do phan giai khong gian thi do phan giai phi thuv-ng bj han chi. Nhung nguyen do tren da thuc diy cac nha khoa hoc nghien cwu phwang phap xwiy nang cao dg phan giai khong gian bing cac thuat toan chuyin dii anh tw do phan giai thap vi do phSn^

giai cao dwa tren chinh cac dw lieu anh thu thap dwac. Bai bao tap trung danh gia mot so phwang phap sieu phan giai (Superresoiution, viit tit SR) phat triin con kha mai me trong thai gian gin day cho muc dich nang cao do phan giai nhim tan dung tit han thong tin anh viin tham VNREDSat-1 cua Viet Nam.

1. Oat van de

He thong ve tinh VNREDSat-1 la he th6ng ve tinh quan sat Trai Oat dhu tidn cua Vl^t Nam Sau khi thiFC hien thanh cong Du- an, Viet Nam hien nay va sau nay se chu dong cung c&p anh vien tham dp phan giai cao cho cac Bo, nganh va cac tinh thanh co nhu c^u SLP dung du" lieu anh vien tham phuc vu phat trien kInh t4 - xa hpi, Cmg pho vai tham hoa thien nhien va bien doi khi hau. De tan dung va irng dung co hieu qua nguon anh vi§n tham dau tien cua Viet Nam cho mue dich giam sat tai nguyen thien nhign va moi tru-ong, b^o ve lanh tho, viec nghien cuu nang eao do phan giai khong gian la can thiet Viec nang cao dp phan giai khong gian theo hu-ang xO ly mpt chu6i cac anh tren chup cung mpt vj tri, co CLing do phan giai de tao ra mpt anh c6 dp phan glal eao hon. Phu-cng phap nay du'p'c gpi la phu-ong phap sieu phan giai - supperresoiution (SR)

Ky thuat quan trpng nhlt trong phu-cng phap SR la phu-cng phap npi suy lam tang hay giam kich thu'oc pixel, SLP dung da anh dp phan giai thap de tai cau true, xu" ly cac han eh6 eua anh don tot hon. Cac thong tin du' thCra chua trong eae anh do phan giai thap co th^

du'p'e du'a vao bang su tjnh tien giua cac subplxel giua chung. Su tinh tien cua subplxel nay eo the xuat hien theo eac doi ttpong hoac theo su chuyen dpng du'p'c dieu khien vi du nhu he thong thu anh tren ve tinh vai quy dao co toe dp va phuang dup-c an djnh trudc.

2. Cac phu'cng phap sieu phan giai

He thong chup anh luon t6n tai cac sai s6 do han che v§ phan cung, anh thu nhgn du^c luon bj suy giam chlit lugng. Vi du, giai han ve dp ma cua 6ng kinh chinh la nguyen nhSn d i n tot hien tugng ma anh, hien tugng nay thuang du'gc mo hinh hoa bai ham Ian truyen theo diem nguon PSF (Point Spread Function). H^n che thai gian phai sang the hien rat ro khi chup nhung doi tugng chuyen dong hoac thiet bj chup chuyen dpng. Han che v l kich

26 TAP CHi KHOA HOC 0 0 OAC VA B A N B 6 SO 20-6/2014,

(2)

thiroc bp cam se dua den anh huong mo do bp cam. Anh huong nay xuat phat tCr viec gia tri pixel anh ehi dupe dya vao mot xung tren mpt don vj dien tieh ma bp cam thu nhan thay vi lay mau xung tCr nhieu bp cam. Han ehe ve mat do cam bi§n tren bp cam se d i n Wi hien tupng rang eua va tCr do lam anh huang tdl dp phan giai khong gian cua tiim anh thu nhan dupe. Nhiing anh huong tren dugc mo phong hoan toan hoac mpt phIn trong eac phuong phap SR khae nhau. Ky thuat sieu phan giai (SR) la phuang phap xay dung Snh dp phan giai cao (HR) tir mpt vai anh do phan giai thlp (LR) do do tang cuang cac thanh phan co tan so cao va ioai bo cac y§u to gay suy giam chat luang anh bang each xi> ly cac anh tir cac bp cam eo dp phan glal thap.

2.1. Mo hinh chup anh

Hinh duoi mo ta mpt he th6ng thu anh dupe mc ta theo [2] vol d^u vao la nhOng anh tu nhien lien tiep vol chat luong dat tol glal han cua he thong.

Hinh 1: He thong tao anh tu' dau thu

"Neu xem X la anh do phan giai cao ky vpng va Y,, la anh thu k dp phan giai thap tir he thong chup anh. Khi do, moi quan he giUa X va Y^ dupe mo ta theo mpt ham 'lexicograph- ical". Gia su, ta thu dupe K anh LR ciia X, a dd anh dp phan giai thap quan he vol anh dp phan giai cao theo cong thuc:

Yk = DkH^FkX + Vk, k =1,2,3...K Trong d6: F^ la ham mo ta chuyen dpng cua anh thu k

Hfc la mo hinh anh huang cua hien tugng ma anh.

D^^ la toan tu down-sampling (toan tu lam min anh).

VJ la ham nhieu.

Ham quan h$ tuyen tinh tren cd the duac the hien theo cong thirc dual dang ma tran;

Dl Hi f i l D2 ff2 fz

DK Hg F,c X +

V2

VK-

Hay Y = IVI.X+V

TAP CHl KHOA HOC DO BAC VA BAN O6 SO 20-6/2014 27

(3)

Trong thuc t§, ma tran M thuang la ma tran thu'a nen ham lien k i t thuo-ng r I t yeu.

Hannu'a, cac thong so D, H, F thuang khdng co dky du ma phai tinh toan dua tren cac s6 lieu thuc te. Do do, viec tinh toan anh tao dp phan giai cao X la khong h§ dan gian. Cac nha khoa hpc da v^ dang co gang tim cac phuang phap to hgp anh dp phSn giai t h i p Y theo nhieu phuang phap toan hoc nham tim ham quan he thuc te kha dgng han.

2.2. Phwofng phap si§u phin giai su* dung mien tin so

Tsai va Huang [1] da su dung moi quan he cOa anh dp phan giai cao vai cac anh dp phan giai thap dugc tjnh t i i n bai cong thuc mien tan so dua tren su djch chuyen tjnh tiin cua cac anh do ph3n giai t h i p va cac thuoc tinh khi> rang cu'a cua chuoi b i i n doi Fourier.

N l u x(t.,,t2) bilu thj cho mpt canh do phan giai cao trong chuoi lien tuc thi se co mot truo'ng K anh tjnh tiin:

^k {^1.*2) = ^(U + Aki, t2 + Ak2), vai k = 1,2 ,K Trong do: Ak^ va Ak2 la tuy blln theo muc dp djch chuyen.

Chuoi lien tuc bien doi Fourier (CFT) cOa mpt canh dugc cho bai X ( Ui,U2) va nhung anh tinh chuyen t u no IS X/^ (Ui,U2). M i l quan he cua chung se dugc xSc djnh theo cong thuc:

^/c{ui.U2) = exp|y2r7<Aki u^ + Ak2 U2)]X(ui,U2)

Anh tjnh tiin dugc lay m i u theo dang xung trong tho'i gian T1 va T2 de co dugc cSc anh dp phSn giai thap:

/[([ni.nz] = x(ni.Ti + Aki, n2.T2 + Akz), vai n, = 0,1,2 ,Ni-1 ; n j = 0,1,2 ,N2-1.

X6t ham Chuoi rai rac cua cac bien doi Fourier (DFTs) cua cac Snh dp phan giai thap Y k[r1,r2]. Bien d i i CFT cOa cac anh tjnh tien c6 moi quan he v6i bien d i i DFTs th6ng qua cac thupe tfnh khu rang cua cua chuoi biin doi Fourier:

Y = <t>.X (•)

Trong do: Via vector K hang 1 opt vol k» la he s6 cua ham DFT yk[r,,r2]

X la vector N,N2 hang 1 cot vai eac gia trj m l u chua biet cua h§ so ham CFT x( t,,t2)

* la ma tran K hang N^Nj ept, la ma tran quan h? giu'a Yva X

Cong thirc (*) djnh nghTa quan h$ tuyen tinh xu^t phat tCr vi?c huo'ng tdl X va sau dd

TAP CHl KHOA HOC DO DAC VA B A N D 6 S6 20-6/2014

(4)

sOr dgng nghjch dao ham DFT de tai c l u true anh. Moi quan he tren sO dgng de tai cau true anh SR vdl gia djnh la khong co nhilu va cac djch chuyen da bilt thSng so day du. Qua trinh giam dO phan giai dugc gia djnh la su dung cac mau xung co anh huang eua hien tugng ma do bp cam viln tham.

Phuang phap miln tan s6 bj giai han doi vai sg- tjnh tien glija cac anh va sg- thilu ro net bat blln (spatially-invariant blurring). Vi the, phuang phap nay khong the su dgng dugc de k i t hgp cac anh chgp a cac goc khae nhau do cac anh co the b| quay theo anh khae gSy ra biin dgng va co khoang each lay mau bat qui tac.

2.3. Phwvng phap sieu phan giai mien khong gian

Hien ton tai rat nhieu phuang phap thong ke linh hoat dugc ung dgng [3], [4] de tai c l u true anh dp phan giai thap. Cong thuc quan he giua anh do phan giai cao va anh do phan giai thap kha giong vai phuang phap phgc hoi anh truyin thing [5], rat nhieu phuang phap thong ke linh hoat dugc ung dgng de tai cau true anh dp phan giai thap. Co the ke d i n eae phuang phap nhu Maximum Likelihood (ML), Maximum a Posteriori (MAP) [6] va phuang phap Projection Onto Convex Sets (POCS) [7].

2.3.1. Ngi suy phuc hdi anh - Phwang phap khong tinh lap

Gia djnh H^ la khong gian tuyin tinh bat bien - Linearly Spatial Invariant (LSI) va dong nhlt d i i v6i mpi khung hinh K, bieu thj khong gian H. Gia su F|( ehi la mo hinh chuyin dong dan giSn nhu tjnh tiin, quay, khi do H va F^ theo [8] thi:

Y^ = D^F^HX+ Vk=Di,FZ, / f = 1 , 2 , ....K,

0§y la ca sa cua phuang phap tinh khong Igp dg-a tren npi suy vS phgc hoi anh. Co 3 budc eg t h i nhu sau:

- Dang ky anh do phan giai thap (khb-p cac anh dp phan giai thap).

- Npi suy khong dong dang d l xac dmh Z.

- Khu mo va nhieu d l co X

Khung einh 66 phan giai thap dugc lien k i t dau tien bai mpt vai nguyen ly phgc hoi anh de dua ra cac subplxel chinh xac. Sg- lien k i t cua cac khung hinh dp phan giai thap sau 66 dugc dgt trong mpt luai dp phan giai cao ma a do npi suy khong dong dang dugc dung de l i p eac pixel bj mat tren anh do phan giai eao de co dugc Z. Sau cung, Z dugc tiin hdnh x6a ma bai phuang phap deconvonlutional truyen thong vai viec loai bo nhieu de tao raX.

Kenren, Peleg, and Brada [9] giai thieu phuang phap tai c l u true anh SR thong qua 2 budc dg-a tren mo hinh chuyen dpng xoln. Ben cgnh do, Gross [10] da dua ra phuang phdp n0i suy khong dong dang de to hgp bp anh dp phan giai thap b^ng each thiet lap nguySn ly tSi ehia mau da kenh. Con Papulis tien hanh phuang phap tai chia mlu su dgng hi$u ung x6a ma. (Xem hinh 2)

Phuang phap npi suy cung dugc lien tyc cai tien, eo the SLP dgng phuang ph^p npi suy tam giic bit quy tac, npi suy lang gllng gan nhat. Uu diem eua phuang phap nay la qua trinh aang ky Anh xu ly de dang tuy nhi§n van eon hgn che trong viec giam hien tugng rang eua trgn dnh.

TAP CHl KHOA HOC 0 0 OAC VA B A N D6 S6 20-6/2014 29

(5)

uwm.wini.(M.Mil4n

Hinh 2: Phuong phap mien khong gian ngi suy - phuc hoi anh 2.3.2. Phuong phap thing ke

Khong nhu phuang phap npi suy - phuc hdi anh, phuang phap thSng ke xem cac buac tai cau anh SR la ngau nhien. Anh dp phan giai cao k i t qua (HR) va eae yeu ta dau vao cija su djch chuyen giira eae anh thu nhan dupe (SR) eo the duac xem nhu la cac biSn nglu nhien. Oat M(v,h) \a ma tran suy giam dugc xac djnh bai vector chuyen dpng v va toan tCr ma h, tai cau true anh SR eo the dupe dua vao dang cua ham Bayesian day du:

X = argmaxPr(X|y)

= argmax \ ?v{Y\X,M(y, Kj) ?T<X) Pr{M(v, h) dv

**(*)**

Chu y rang, X va M(v,h) va dupe thdng ke dpc lap theo [11]. 0 day, Pr('K,M(v,h)) la dii' lieu thong tin anh , Pr(X) la han sal am ehi anh dp phan giai eao mong mu6n, Pr(M(v,h)) la thuat ngu' the hien sy thong ke cua chuyen dpng. V trong cong thirc tren thuang la chi dp nhieu, gia djnh la bang khong va vai mpt vector Gaussian rcng. Thep do:

Fr(r|x,M(i;,/i))cc e j ; j , | - | | y - M ( i 7 , / i ) ; f | p |

Pr(X) dupe xac djnh bang each sir dung ham phan ph6i Gibbs theo ham mu dang;

P r ( « = iexpC-«4(;f)}

Trong do: A(X) la mpt ham kha nang khong am; Z la y§u to chuan hoa.

Ham Bayesian theo cong thirc (*) rat phire tap va rat kho de co the thuc hien danh gia thdng ke khi ve tinh chuyen dpng.

2.3.3. Phuxyng phap thiit lap nguyen ly phuc hdi anh

TAP CHi KHOA HOC DO BAC VA B A N O 6 SO 20-6/2014

(6)

Ben canh cac phuang phap tiep can toi uu hda tip viec sCr dung bi4n nglu nhien nhu da thao luan a tren, mpt chieu huang tiep c^n khae thong qua Phep chieu tren tap loi - Projection onto Convex Sets (POCS)) dugc giai thieu trong [12]. Phuang phap POCS giai quyet van de eOa anh dp phan giai thip bang each xay dung mpt tap loi vai nhieu giai han trong do bao gom ca yeu cau ve anh, yftu cau nay dupe xem nhu mot diem loi. Nhung djnh nghTa ve tap loi tuang doi linh hoat va co the ket hgp nhi§u loai dieu kien hoac dat truoc, tham chi la nhD'ng gid'i han phi tuyen va khong co tham s6.

Tich hgp du lieu vol nhtrng dieu kiSn tai cau true eo the dugc mo hinh hoa dud'i dang tap loi K:

C/c ={X I ||D»H;^p<-- Yk II'< a', 1 < A < K}

Trong do, t5n tai dieu kien lam mjn va di Su kien bien dp. Ky thuat POCS dua ra thuat tpan tim cac diem giao cua tap X:

Xk+1=PMPM-1 P2PlXk

Trong do, XQ la gia trj dat ban dau; Pj la toan tu chiiu - toan tu chieu nay ehilu mot diim len mat t|ip loi khep kin C,.

NhD'ng uu diem cua ky thuat POCS la sg- dan gian va eo the kit hgp vdi mpi g\6\ hgn cung nhu dieu kien cho truae ma phuang phap tiep can nglu nhien khong cho ph6p. Tuy nhien, POCS co muc dp tinh toan rat nang va hoi tg cham. Giai phap cua POCS cung khong hoan hao va phg thupe nhieu vao viec dat thong so ban dau. Cac phuang phap POCS cung gia djnh trudc cac thong so ve chuyin dpng va su ma hp thong. No khong t h i ude tinh thong qua nhung th^nh phan da dugc dang ky truac va anh dp phan giai cao nhu phuang phap tilp c i n nglu nhi^n.

3. Oe xuat phu'cng phap sieu phan giai (SR) trong XCP ly anh VNREDSAT-1 OCrng tren phuang dien tong t h i thi eo hai phuang phap chinh la phuang phap dg-a tren miln tan so va mien khong gian. Phuang phap mien tan so thuang su dgng chuoi Fourier d l mo phong djch chuyen dong thai da anh ban dau, Tuy nhien, do tinh chat phuc tap cua djch chuyen da anh nen h^m mo phong thuang khong du tham so do do viec giai bai toan rIt kh6 khan va mo hinh khong dugc chat che. Do do phuang phap nay it dugc phat triin nSng cao. Phuang phcip su dgng mien khong gian vai dac diem la uac lugng cac thong s i tru6c dg'a tren cac thong tin v l anh va yeu cau v l anh do phan giai eao d i u ra da giai quylt v I n de khong du tham so eua phuang phap miln tan so. Day la huang di dugc phat triin md" rpng mgnh me trong nghien euu sieu phan giai. Tu truae d i n nay, phuang phap mien khong gian co nhilu phuang phap nhung do la su phat trien noi tilp nhau, phuang ph^p giSi quylt nhOng van d l ton tai eua phuang phap truae. Xet ve ky thuat thi e6 hai hud-ng di ehinh d6 lei phuang phap su dung bien ngau nhien uac lugng thong qua h^m x^c xult va phuang ph^p ude lugng eae biin thong qua xac djnh tap III bing phep chieu t$p l l i v^i c^c diiu kipn bi§n dugc xac djnh truae. Ngoai ra, phuang phap k i t hgp e^ hai phuang ph^p tren cung dugc phat triin va nhD'ng k i t qua nghien euu ban dau mang Igi k i t qu3 tuang d i i kha quan.

Vi v§y, dg'a tren nguySn ly phuang phap su dgng mien khong gian thi d l xu ly 5nh VNREDSat-1 cho mgc dich nang cao, de xuat su dgng eae buac ca ban sau:

TAP CHi KHOA HOC 0 0 DAC VA BAN O 6 S 6 20-6/2014

(7)

- Thu nhan da anh tren cung mpt (01) vi tri;

- Mo phfing m6i quan he giira cac anh thu nhan dugc, thuang la thdng qua cac m6 hinh chuyin dpng giiJa eae anh vai doi tugng chyp;

- Phan chia Pixel, thuang chia mpt (01) pixel ban dau thanh bon (04) sub-pixels;

- Lap ma tran cac gia trj pixel ban dau de xac djnh cac gia tri pixel cua anh dp phan glal eao ket qua;

- Tai cau true anh.

Qua trinh tinh toan thuang khong phan ehia ro ret thanh cac buac nhu tren ma cd the thuc hien tirng buac hoac thilt lap mpt mo hinh thong nhlt k i t hgp t i t ca eae budc tren.

4. Ket lu$n

NhD'ng van de tro' ngai Icm doi vol phuang phap SR xult phat tip qua trinh khap anh, viec tinh toan rat nang ciia qua trinh tinh toan tham so mo hinh phuc hdi anh.... Khftp anh quyet djnh chat lugng l l y mau khdng gian b6 sung vao qua trinh phan chia pixel va tai elu tnJc. KJ thuat kh6p anh e i n phai h i t sue luu tam vol anh viln tham tai nhung viing dia hinh bien doi phire tap. Nhilu gay ra bai eac sai so trong qua trinh khap anh anh \&n han hieu irng mg do npi suy. Dp chinh xac phan chia cac subplxel quylt djnh gia trj pixel anh dp phan giai cao. Do dd, de phu hgp vo'i viec tai cau trOe anh SR thi viec khgp anh LR cd the glal quylt eiing vdi trugt xoan anh truo'c khi tai cau trile anh d l thu dugc anh e6 do phan giai khdng gian cao han.

Mpt han o h i irng dung thye t l eCia phuang phap tai e l u trCie SR la tinh toan doi hdi thao tac vgi eae ma tr$n h i t sire ton b0 nho may tInhTren thyc t l , vo'i anh vien tham thi khil lugng pixel anh Ion chinh la van de dang quan tam. Tuy nhien, cac nha khoa hpc cung da de xuat nhilu bien phap tinh tcan nham giam bp nha dong tho'i tang toe dp tinh toan On thong qua viec luge bd cac thanh phan khong can thiet, xay dyng mo hinh ehat che tir dau d l loai trir bt/t cac thflng so da bilt vao tinh toan.

D i n nay, de dua ra dugc han c h i cho tat ca eae ky thuat SR la rat kho. Viec sCr dgng phuang phap SR cho anh VNREDSat-1 la mpt nhiem vy phi>c t?p bao gom nhieu budc tinh toan phg thupe Ian nhau.

sai bao la k i t qua cua de tai "Nghien ciru phuang phap xir ly nang cao ehat lygng anh ve tinh VNREDSat-1", ma so VT/UD-01/13-15, thupe Chuang trinh khoa hpc cong nghe dpc lap cap Nha nuac ve Ccng nghp vu trg ma sp KHCN-VT/12-15.0

Tai lieu tham khao

[1]. R. Y. Tsai and T. S. Huang. Multipleframe image restoration and reglstration.ln Advances in Computer Vision and Image Processing, pages317-339. Greenwich, CT: JAI Press Inc., 1984.

[2]. IVI. Elad S. Farsiu, D. Robinson and P. Milanfar. Advances and chal-lenges in super- resolution. International Journal of Imaing Systems andTechnology, 14(2):47-57, 2004.

[3]. Sean Borman and Robert L. Stevenson. Super-resolution from image sequences- A review. In Proceedings of the 1998 IVIidwest Symposium on Circuits and Systems, pages 374-378, 1998.

32 TAP CHl KHOA HOC DO BAC VA B A N D 6 S6 20-6/2014

(8)

[4]. S. Baker and T. Kanade. Limits on super-resolution and how to break them. IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(9):1167-1183, 2002.

[5]. M. Elad and A. Feuer. Restoration of single super-resolution image from several blurred, noisy and down-sampled measured images. IEEE Transaction on Image Processing, 6(12):1646-1658,1997.

[6]. R. C. Hardie, K. J. Bamard, and E. E. Amnstrong. Join MAP registration and high resolution image estimation using a sequence of undersampled images. IEEE Transactions on Image Processing, 6(12):1621-1633, 1997.

[7]. H. Stark and P. Oskoui. The method of convex projections and its application to image reconstruction in computerized tomography Control and Applications, 1989.

Proceedings. ICCON '89. IEEE International Conference on. tr 241-244, 1989.

[8]. S. Farsiu, D. Robinson, M. Elad, and P. Milanfar. Fast and robust multi-frame super- resolution. IEEE Transaction on Image Processing, 13(10):1327-1344, 2004.

[9]. D. Keren, S. Peleg, and R. Brada. Image sequence enhancement using subplxel dis- placements. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 742-746, 1988.

[10]. H. Ur and D. Gross. Improved resolution from subplxel shitted pictures. CVGIP:

Graphical Models and Image Processing, 54(2):181-186, 1992.

[11]. R. C. Hardie, K. J. Barnard, and E. E. Armstrong. Join MAP registration and high resolution image estimation using a sequence of undersampled images. IEEE Transactions on Image Processing, 6(12):1621-1633, 1997.

[12]. D. C. Youla and H. Webb. Image registration by the method of eonvexprojections:

Part 1-thoery. IEEE Transactions on Medical lmaging,1(2):81-94, 1982.0

Summary

Improving resolution of remote sensing images by super resolution methods Dn Le Quoc Hung, Eng. Truong Thi Tuyet

Vietnam Remote Sensing Center

The method for improving remote sensing image spatial resolution is the target of the top research scientists in the field of remote sensing. Up to now, the way to improve image spatial resolution by upgrading and manufacture of new advanced sensors that allow tak- ing high spatial resolution images has enormous costs. In addition, spectral resolution was often limited while enhanced spatial resolution. These reasons prompted scientists to dis- cover the method for improving spatial resolution by transformed algorithms to high resolution images based on received images. This article focuses on assessing some super- resolution methods which develop so far. The result will be scientific basis for the purpose of enhancing resolution to better use of VNREDSat-1 data.O

Ngiy nh$n bil: 10/3/2014.

TAP CHl KHOA HOC DO OAC VA BAN O6 S6 20-6/2014