Master informatique spécialité IMA Cours N°2 Traitement d’images
cours du 2 novembre 2010
Catherine Vénien-Bryan!
Equipe “Assemblages macromoléculaires” !
Institut de minéralogie et de physique des milieux condensés (IMPMC)! CNRS UMR 7590, Universités Paris 6 et Paris 7, IPGP!
140 Rue de Lourmel, 75015 Paris! Tél.: 01 44 27 72 05!
E-mail: catherine.venien@impmc.upmc.fr! catherine.venien@bioch.ox.ac.uk!
URL: http://www.impmc.jussieu.fr
!"#$%%&&&'()*+!'*,'-+'./%-0#0)12%)342,'-0#5#-62)4789:;
Traitement d’images
Particules isolées, protéines macromolécules, petits virus (haute résolution)
•" Numérisation de l’image, cameras CCD
•" Sélection des particules et normalisation du contraste
•" Analyse dans les 2dimensions:
•" Alignement
•" Classification des images
•" Création du modèle dans les 3 dimensions et rafinement
•" Méthode des séries coniques aléatoires
•" Lignes communes (transformée de Radon)
•" Rafinement
Cellules entières, bactéries virus (moyenne résolution)
•" Tomographie électronique cellulaire
•" Cemovis (cryo-electron-microscopy of vitreous sections)
!!"#$%&'(%#
Successfully used in many TEM applications that are difficult or impossible to carry out without an online digital imaging system, (microscope autotuning, automated electron tomography,
electron holography, protein electron crystallography and telemicroscopy)
<;=>*-4;4?3-@)+;>-362;
<;A)32->)1?;
<" ;A*&;B2C2B;*D;3*)02;
EB2+1>*3;@)+>*0+*#?$;
<" ;F*;*#23)36;*D;1!2;+-@2>-;C-+..@;D*>;GB@;(*,;2,+!-362;
H+-3;(2;-;@-I*>;0*.>+2;*D;&-12>;+*31-@)3-310;)3;-;
@)+>*0+*#2;+*B.@3J;
<;K@@24)-12;L.-3M1-MC2;D224(-+/;-(*.1;1!2;L.-B)1?;*D;
(*1!;1!2;0-@#B2;-34;1!2;4-1-;(2)36;+*BB2+124;>-1!2>;1!-3;
!-C)36;1*;&-)1;1*;
42C2B*#;-34;1!23;1*;4)6)MN2;1!2;GB@;
O)B@;)0;0MBB;.024;)3;PEQ;-##B)+-M*30;
R" ;O)B@;
R" ;S-M0D-+1*>?;>2+*>4)36;.342>;-;&)42;>-362;*D;2B2+1>*3;232>6)20;
R" ;TTU;
R" ;U)V+.B1?;*D;.0)36;TTU0;D*>;)@-6)36;&)1!;!)6!;232>6?;2B2+1>*30;
R" ;A)@)124;3.@(2>;*D;#),2B0;*W2>24;(?;1!2;+.>>231B?;-C-)B-(B2;0?012@0X;-0;
+*@#->24;&)1!;1!-1;*(1-)324;(?;4)6)MN)36;-;PEQ;326-MC2;H0@-BB;G2B4;*D;C)2&J;
R" ;TTU;D*>;B*&<>20*B.M*3;01.4)20;H!)6!;SFY;-1;B*&;D>2L.23+)20X;(2"2>;+*31>-01JX;GB@;D*>;
!)6!<>20*B.M*3;01.4)20;H!)6!;SFY;-1;!)6!;D>2L.23+)20X;421-)B0;(2"2>;>21>)2C24J;
Zeiss SCAI
Nikon Coolscan
)*+&#,$%--'(,#
Geometric accuracy: Scan of
regular grid at 0 and 90° orientations DQE: digital quantum efficiency - (SNR)
2(signal to noise ratio) of scan divided by (SNR)
2of image.
MTF: modulation transfer function - like the CTF, amplitude of signal
transmitted by scanner as a function of resolution, but no phase changes
Criteria for evaluating film scanners
resolution
MTF
Digital images
Sampling and grey level
Discrétisation spatiale et quantification Image Digitization
Original image Image sampled at low resolution
Grey level resolution:
one bit can code for 2 states:
“0” or “1”
Byte (octet): a string of 8 bits can code for 2
8= 256
different states Pixel size
The image must be divided up into pixels (sampled) at a spacing at least twice as fine as the finest detail (highest frequency) to be analysed.
(in practice, 3-4x as fine).
Each picture element
stored in the computer,
with its own grey level, is
called a pixel.
Digital images Sampling-1
Q*>2;)0;(2"2>ZZZ;
Digital images Sampling- spatial resolution Théorème de Nyquist-Shannon
Nyquist-Shannon frequency: the sampling frequency is twice the highest frequency term being represented.
If the same sampling frequency is used for a higher frequency wave, the sample points do not follow the oscillations. The features will not be correctly represented in the image.
Remember that the image can be represented as the sum of a series of sinusoidal waves. The highest frequency wave term present defines the resolution limit. When an image is digitised, the wave components are sampled at an interval defined by the scanner step size.
•" In practice, it is necessary to sample at 3-4x the
resolution, to avoid numerical rounding errors.
Digital images Scanning – number of grey levels-
144 72 16
8 4 2
Sampling - grey scale resolution-
Film scanners generally scan into 8 bits (256 grey levels) or more.
But if they are not set up correctly it is quite possible to end up with a digital image represented by far fewer grey levels. This can lead to degradation of the image and loss of resolution.
Scaled into 256 grey levels
5 grey levels
3 grey levels
Statistical noise in low electron dose images
Electron beam
Photographic film or detector
Random pattern of incident
electrons
N = 100 electrons per unit area
N = 1000 electrons per unit area electron dose = average (N) ± ! N
! N / N = 10% ! N / N = 3.3%
The signal to noise ratio improves by a factor of !N as the electron dose increases.
However, for beam-sensitive specimens the dose must be kept low to avoid radiation damage to the specimen. Electrons can transfer energy to the specimen, breaking bonds and causing mass loss in biological molecules.
For analysis, we assume that the image is the sum of the structure information plus random noise.
The operator needs to find a compromise between the SNR which increase with the number of electrons and the damage cause by them
Noise reduction by averaging
Averages of 2 5 10 25 200 images Raw
images
Variance
For a stack of aligned images, the variance can be calculated for each pixel, to give a map of variations between the images in the data set.
This can help to assess the reliability of features seen on the average image, and can reveal if images of different structures are mixed up in the same data set.
The variance is determined for each pixel as the difference between the pixel value in a given image and the average value of that pixel in all the images. This difference is squared and the sum of these squares is calculated for all the images in the stack.
Variance = [1/(N-1)] " [P
i(r
j) - P
av(r
j)]
2where Pi(r
j) is the value of pixel j in image i and P
av(r
j) is the average value of pixel j in all the images, for a set of N images.
i,,j
.-%+/,'#01*&%2',#0%-,#+',#3"#
45+'$67-#0',#8%(6$9+',#:
!2@*+?-3)3;!"#$%&'()*+*;#
Normalisation du contraste
Soustraction de la moyenne et division par l’écart type
Alignement et fonctions de corrélation Corrélation croisée 2D
[;7;Y2D; =;7;K@-62;
and =
=;201;1>-30B-1\;21;];+!-L.2;#*0)M*3;42;=;;
23;^;21;_;.3;3*.C2-.;+*2V+)231;42;+*>>\B-M*3;;
201;+-B+.B\';
A2;>\0.B1-1;G3-B;201;.32;+->12;()4)@230)*332BB2;4*31;B-;1-)BB2;;
+*>>20#*34;];+2BB2;420;)@-620;4`*>)6)32;21;L.);#*00a42;.3;;
#)+;42;+*>>\B-M*3;L.);+*>>20#*34;];B-;1>-30B-M*3;42;=;#*.>;;
B-L.2BB2;B20;42.,;4)0L.20;02;0.#2>#*0231;#->D-)12@231';
A-;#*0)M*3;4.;#)+;#->;>-##*>1;];B`*>)6)32;H+231>2;42;B-;+->12J;
+*>>20#*34;-.;C2+12.>;42;1>-30B-M*3;];-##B)L.2>;#*.>;
>2+231>2>;B`)@-62;=;#->;>-##*>1;];B`)@-62;[;H,;bcd;X;?;<cdJ' ;;
? ;
^;
ee;;/H)XIJ;';DH)b3XIb@J;
/H)XIJ; DH)XIJ;
^;1>-30B-M*3;3;7;9;;
_;1>-30B-M*3;@;7;9;
Alignement et fonctions de corrélation
Fonction d’Auto-Corrélation = FAC
T`201;B-;+*>>\B-M*3;+>*)0\2;4`.32;)@-62;-C2+;2BB2<@f@2;
K@-62; [TO;
Corrélation croisée angulaire
g32;)@-62;=;H0.##*0\2;+231>\2J;201;#B-+\2;0.>;.32;;
)@-62;42;>\D\>23+2X;21;#*.>;+!-L.2;>*1-M*3;42;=X;*3;@20.>2;
B2.>;+*2V+)231;42;+*>>\B-M*3';
Alignement et fonctions de corrélation
>2D;
=;
Y*1-M*3;-36B2;
+*>>2B-M*3;
A-;+*>>\B-M*3;201;+-B+.B\2;0.>;.32;0\>)2;42;;
>-?*30;21;B-;C-B2.>;6B*(-B2;201;)30+>)12;23;;
D*3+M*3;42;B`-36B2;42;>*1-M*3';
Exemple d’alignement sur une référence
Y\D\>23+2;;;;;;;;;;;;)@-62;=;
O[T;>2D;;;;;;;;;;;;;;;O[T;=;
T*>>\B-M*3;+>*)0\2;-36.B-)>2;420;O[T0;
Q-)0;+*@@2;B20;O[T0;0*31;420;D*3+M*30;#-)>20X;
)B;?;-;.32;-@()6.h1\;42;ci9j;0.>;B`-36B2;42;
>*1-M*3;L.`*3;1>*.C2;23;+-B+.B-31;B2.>;
+*>>\B-M*3;+>*)0\2;-36.B-)>2';
k3;+-B+.B2;B20;O[T0;420;)@-620;
#*.>;*(123)>;.32;>2#>\0231-M*3;;
+231>\2;21;>2l\1-31;#->M2BB2@231;
B-;01>.+1.>2;420;#->M+.B20';
Alignement et fonctions de corrélation
O@-,;7;9'mc; O@-,;7;9'n8;
[B)632@231;21;D*3+M*3;42;+*>>\B-M*30;
[B)632@231;0.>;.32;>\D\>23+2<:;
D’autres méthodes d’alignement ont été proposées pour minimiser l’influence de l’image de référence.
"Reference free" iterative alignment (Penczek et al., 1992) :
1) Ici deux images sont prises au hasard, alignées et leur moyenne est alors utilisée comme nouvelle référence pour aligner une troisième image. Le processus se reproduit itérativement jusqu’à ce que toutes les images
soient alignées.
2) Pour minimiser l’influence de l’ordre dans lequel les images ont été choisies pendant l’alignement, on repart ensuite à l’envers, en réalignant la première image et en la réalignant sur (Moyenne totale - l’image 1). Puis la seconde image est réalignée sur (Moyenne totale - l’image 2), etc …
3) Le processus entier est recommencé à nouveau sur les images issues de ce premier cycle d’alignement (étapes 1 et 2), jusqu’à ce qu’aucune amélioration ne soit constatée d’un cycle au suivant.
"multi-reference alignment" et d’autres méthodes d’alignement utilisent la
classification en parallèle du processus d’alignement.
<='&8+'#0'#$'->(%2'#'>#01%+*2-'&'->#?#@'&7$/%-*-AB#
<='&8+'#0'#$'->(%2'#'>#01%+*2-'&'->#?#@'&7$/%-*-A3#
!+%,,*C$%67-#3"AB#
Classification 2D-2
Brétaudière JP and Frank J (1986) Reconstitution of molecule images analyzed by correspondence analysis: A tool for structural interpretation.
J. Microsc. 144, 1-14.
!+%,,*C$%67-#3"AD#
Image No1
Image No80
Pixel 1 Pixel 2754
x
ij!+%,,*C$%67-#3"AE#
Q-1>)+2;
Méthode de la dragée
Espace à 2754 dimensions
#
:;#
o;#
c;[,2;D-+1*>)2B;c;
[,2;D-+1*>)2B;:;
[,2;D-+1*>)2B;o;
A-;4)-6*3-B)0-M*3;42;B-;@-1>)+2;+->>\2;
T = X'X
;#2>@21;42;4\12>@)32>;B-;#B.0;6>-342;4)>2+M*3;4`2,1230)*3;42;3*0;4*33\20;H3.-62;42;#*)310J;4-30;B`20#-+2;@.BM<4)@230)*332B';T2"2;4)>2+M*3;4`2,1230)*3;+*>>20#*34;];B-;
#B.0;6>-342;p;C->)-M*3;q;*.;p;1234-3+2;q;-.;02)3;42;3*0;4*33\20';T`201;B`-,2;D-+1*>)2B;Fjc;4*31;B`-@#B)1.42;201;
+->-+1\>)0\;#->;B-;C-B2.>;#>*#>2;
#
c';k3;2W2+1.2;-B*>0;.3;+!-362@231;42;>2#a>2;#*.>;4\12>@)32>;B-;#*0)M*3;42;+!-+.32; 42; 3*0; 4*33\20; H)@-620J; #->; >-##*>1; ]; +21; -,2; D-+1*>)2B'; r.)0X; *3; >2+!2>+!2; B-; 02+*342; #B.0; 6>-342;
4)>2+M*3;4`2,1230)*3;p;*>1!*6*3-B2;q;];B-;#>2@)a>2;#*.>;4\G3)>;B`-,2;D-+1*>)2B;Fj:;+->-+1\>)0\;#->;.32;-@#B)1.42;42;
#
:'; A2; D-)1; L.2; B20; -,20; D-+1*>)2B0; Fjc; 21; :; 0*)231; *>1!*6*3-.,X; )34)L.2; L.`)B0; +->-+1\>)0231; 420; C->)-M*30;)34\#234-3120;H3*3<+*>>\B\20J';k3;2,#>)@2;-)30);#->;*>4>2;4\+>*)00-31;1*.120;B20;C->)-3+20;)34\#234-3120;42;3*0;
4*33\20;0.>;B20;-,20;D-+1*>)2B0;FjcX;:X;oX;s;X21+;t;;;
!+%,,*C$%67-#3"AF#
!(5%67-#019-'#&%>(*$'#%#3GFH#0*&'-,*7-,#
Classification 2D
a) Méthodes par partition: e.g. "Moving seeds" method
Diday E (1971) La méthode des nuées dynamiques. Rev. Stat. Appl. 19, 19-34.
Deux images prises au hasard servent de centres d’agrégation pour la partition. Les
centres de gravité de chaque classe servent de nouveaux centres d’agrégation pour un
nouveau cycle de partition. Arrêt lorsque les centres d’agrégation ne bougent plus d’un
cycle à l’autre, ou après un nombre déterminé d’itérations.
Classification 2D
b) Classification Ascendante Hiérarchique
c; :;
o;
s;
d;
c; :;
o;
s;
d; c; o;
8;
8;
d;
:;
n;
n;
s;
i;
i;
I'J'('-$',#
S!-)/!;P'X;u-*X;v'X;=-,12>Xw'P'X;[01.>)-0X;O'x'X;=*)0021X;F'X;A2)1!X;['X;-34;O>-3/X;x';H:99iJ;SrKUEY;)@-62;#>*+200)36;
D*>;0)36B2<#->M+B2;>2+*301>.+M*3;*D;()*B*6)+-B;@-+>*@*B2+.B20;D>*@;2B2+1>*3;@)+>*6>-#!0;,*-./"&
0/(-(1(#2&o;Hc:J;cmsc<cmns';
T->>-6!2>;=X;y)002(2>1!;FX;y>)26@-3;UX;Q)BB)6-3;Y[X;r*"2>;TSX;r.B*/-0;xX;Y2)B2)3;[;H:999J;A26)3*3$;-3;
-.1*@-124;0?012@;D*>;-+L.)0)M*3;*D;)@-620;D>*@;C)1>2*.0;)+2;0#2+)@230';34&5-/.1-4&6$(#(78;co:X;oo<sd';;
r*"2>;TSX;T!.;vX;O>2?;=X;u>223;TX;y)002(2>1!;FX;Q-4423;PxX;Q)BB2>;yAX;F-!>01241;yX;r.B*/-0;xX;Y2)B2)3;[X;P+!236;
UX;w2(2>;UX;T->>-6!2>;=';HcmmmJ;O>-3/X;x';HcmndJ';[C2>-6)36;*D;B*&;2,#*0.>2;2B2+1>*3;@)+>*6>-#!0;*D;
3*3#2>)*4)+;*(I2+104&9#-/*)$1/(21('8;cX;cdm<c8:';
O>-3/X;x'X;-34;[B<[B)X;A';HcmndJ';S)63-B<1*<3*)02;>-M*;*D;2B2+1>*3;@)+>*6>-#!0;*(1-)324;(?;+>*00<+*>>2B-M*3';
,*-./";:d8X;on8<oni';
O>-3/;x';Hcmm8J;P!>22<4)@230)*3-B;@)+>*0+*#?;*D;@-+>*@*B2+.B->;-002@(B)20';:1*;")$1&0/"22X;S-3;U)26*';
O>-3/;x';Hcmm9J;TB-00)G+-M*3;*D;@-+>*@*B2+.B->;-002@(B)20;01.4)24;-0;z0)36B2;#->M+B20z';{.->1';Y2C';=)*#!?0';:oX;
:ic<o:m';
v2342>0*3X;Y'X;-34;uB-202>X;Y'Q';HcmidJ';{.-3M1-MC2;-3-B?0)0;*D;)@-62;+*31>-01;)3;2B2+1>*3;@)+>*6>-#!0;*D;
(2-@<0230)MC2;+>?01-B0';9#-/*)$1/(21('8;c8X;com<cd9';
v2342>0*3X;Y';HcmmdJ';P!2;#*123M-B;-34;B)@)1-M*30;*D;32.1>*30X;2B2+1>*30;-34;^<>-?0;D*>;-1*@)+;>20*B.M*3;
@)+>*0+*#?;*D;.301-)324;()*B*6)+-B;@*B2+.B20';{.->1';Y2C';=)*#!?0';:iX;cnc<cmo';
S-,1*3X;w'k';HcmmsJ;[++.>-12;-B)63@231;*D;0210;*D;)@-620;D*>;0.#2><>20*BC)36;-##B)+-M*30X;34&<$1/(21';cnsX;
8c<8i';
C-3;v22BX;Q';HcmimJ';TB-00)G+-M*3;*D;C2>?;B->62;2B2+1>*3;@)+>*0+*#)+-B;)@-62;4-1-;0210';='+>;H3X;ccs<c:8;;
S\>)2;42;#>*I2+M*30;
)3+B)3\20;
T-B+.B;4`.3;C*B.@2;
42;>2+*301>.+M*3;oU;;
#->;>\1>*#>*I2+M*3;
K(*-$*8'#0'#+%#>7&72(%8@*'#&7+5$9+%*('#:B; ;;
2 ; 5
3 4
6 1 7
E0#-+2;42;O*.>)2>;
1 2 3 4 5 6 7
P!\*>a@2;42;B-;02+M*3;+231>-B2$;
E3;20#-+2;>\+)#>*L.2X;1*.12;#>*I2+M*3;()<
4)@230)*332BB2; 4`.3; *(I21; +*>>20#*34; ];
.32; 02+M*3; +231>-B2; 4-30; B-; 1>-30D*>@\2;
42; O*.>)2>; oU; 42; B`*(I21'; T!-L.2; 02+M*3;
+231>-B2; 201; *>)231\2; #2>#234)+.B-)>2@231;
]; B-; 4)>2+M*3; 42; #>*I2+M*3; H4)>2+M*3; 4.;
D-)0+2-.;4`\B2+1>*30J';;
K(*-$*8'#0'#+%#>7&72(%8@*'#
:,5(*',#*-$+*-5',;#:3; ;;
X y
Z
$ % & ;
X
Y Z
A20;+*3C23M*30;SrKUEY;
+*3+2>3-31;B20;-36B20;2.B\>)230';
$" = phi = rotation autour de Z '
% = theta = rotation autour de Y '
& = psi = rotation autour de Z
!7->(%*->'#0'#+%#$(/7L<M#,9(#7NO'>,#N*7+72*P9',#Q#
P>-C-)BB2>;];D-)(B2;4*02;4z\B2+1>*30;Hc92
<%|
:J;
S2.B2@231;c;H*.;:J;#>)02H0J;42;C.2H0J;;#->;N*32;
T*BB2+12>;F;)@-620;];D-)(B2;>-##*>1;0)63-B%(>.)1;
[C2+;.32;D*>12;0*.0<D*+-B)0-M*3;H+*31>-012J;
T-B+.B2>;420;@*?23320;3.@\>)L.20;:U;*.;oU;
T*>>)62>;B-;D*3+M*3;42;1>-30D2>1;42;+*31>-012;HTPOJ;
gMB)02>;B20;0?@\1>)20;21%*.;4)0#*0)M*3;420;#->M+.B20$;
S?@\1>)2;!\B)+*h4-B2X;)+*0-\4>-B2X;>\02-.,;:UX;
*.;#-0;42;0?@\1>)2';
R%#('$7-,>(9$67-#019-#&70S+'#8(5+*&*-%*('#D"#
L’acquisition des séries coniques aléatoires
Les méthodes de reconstructions
0°
45°
1 ; 2 ;
5 3 4
7 6
1 2 3 4 5 6 7 8 8
r->M+.B20;)0*B\20$;[C2+;02.B2@231;42.,;#>)020;42;C.20;*3;#2.1;;
+-B+.B2>;.3;C*B.@2;42;>2+*301>.+M*3;
S\>)20;+*3)L.20;-B\-1*)>20;
9j;
sdj;
45° ; 0°
9j;
sdj;
<='&8+'#0'#('$7-,>(9$67-#T#8%(6(#0',#*&%2',#'=85(*&'->%+',;
:v\@*+?-3)32;4;`!"#$%&'()*+*?&0?@\1>)2;U
dJ;
}.20;42;+~1\;(>.120; Qf@20;)@-620;-B)63\20;
[B)632@231;
U\12>@)3-M*3;
42;B;`-36B2;H()'
TB-00)G+-M*3;-.1*@-ML.2;!;d;C.20;
T-B+.B;4`.3;C*B.@2;#->;+B-002;];#->M>;420;)@-620;)3+B)3\20;
[002@(B-62;420;C*B.@20;-B)63\0;;
S.>D-+2;2,12>32; T-C)1\;+231>-B2;
v\@*+?-3)32;4`!"#$%&'()*+*&
.#('>'-*(#,9(#+1%88(7$@'#0',#,5(*'#$7-*P9',#%+5%>7*(',#
c; T2"2; @\1!*42; 3\+200)12; 42; +*BB2+12>; 420; )@-620; -C2+; 21; 0-30; )3+B)3-)0*3; 4-30; B2;
@)+>*0+*#2';A-;@)+>*0+*#)2;-;.3;D-)(B2;>2342@231;+->;)B;?;-;420;2W210;42;+!->62;21;
42;0.#2>#*0)M*3;#->M2BB2;420;#->M+.B20;L.);32;0*31;#-0;@-•1>)0-(B20';
:;A-;12+!3)L.2;D*3+M*332;@f@2;0.>;B20;)@-620;];D-)(B2;>-##*>1;0)63-B;%;(>.)1;42;D*34;
! ;*3;1>-C-)BB2;4*3+;0.>;B20;)@-620;(>.120';
o;KB;?;-;1*.I*.>0;.3;->1\D-+1;4.;+~32;@-3L.-31;L.);#2.1;f1>2;0.##>)@\;0);*3;>26>*.#2;
420; C*B.@20; +*>>20#*34-31; ]; 420; C.20; 4)W\>23120; 42; B-; #->M+.B20X; *.; 0); B-;
#->M+.B20;-;420;0?@@\1>)20;)312>320;D-C*>-(B20';H2,2@#B2$;#->M+.B2;+?B)34>)L.2;];
0?@\1>)2;U
d$;A20;C.20;42;4200.0;#>*4.)0231;.3;C*B.@20;-3)0*1>*#2;-C2+;.3;->1\D-+1;
4.;+~32;@-3L.-31;-B*>0;L.2;B20;C.20;42;+~1\;#>*4.)0231;.3;C*B.@2;)0*1>*#2J';
R',#&5>@70',#01%U-'&'->#0'#+1%+*2-'&'->#'>#0'#+%#('$7-,>(9$67-#
Y2L.)>2;-;>2D2>23+2;oU;@*42B';
Y2-B;0#-+2
;$;;S!-)/!;P'X;u-*X;v'X;=-,12>Xw'P'X;[01.>)-0X;O'x'X;=*)0021X;F'X;A2)1!X;['X;-34;O>-3/X;x';H:99iJ;SrKUEY;)@-62;#>*+200)36;D*>;
0)36B2<#->M+B2;>2+*301>.+M*3;*D;()*B*6)+-B;@-+>*@*B2+.B20;D>*@;2B2+1>*3;@)+>*6>-#!0;,*-./"&0/(-(1(#2&o;Hc:J;
cmsc<cmns';SrKUEY;0*€&->2;
r23+N2/X;r'['X;u>-00.++)X;Y'['X;O>-3/X;x';HcmmsJ;P!2;>)(*0*@2;-1;)@#>*C24;>20*B.M*3$;32&;12+!3)L.20;D*>;@2>6)36;
-34;*>)231-M*3;>2G32@231;)3;oU;+>?*<2B2+1>*3;@)+>*0+*#?;*D;()*B*6)+-B;#->M+B20';9#-/*)$1/(21('8;FD$;:dc•:n9';
S#)42>;0*€&->2;
O*.>)2>;1>-30D*>@
$;u>)6*>)2WX;F';HcmmiJ';P!>22<4)@230)*3-B;01>.+1.>2;*D;(*C)32;F[Uv$;.()L.)3*32;*,)4*>24.+1-02;H+*@#B2,;KJ;-1;::;[;
)3;)+2';34&<(#4&6$(#4;3GGQ;c9oo•c9s8';
x*3)+X;S'X;S*>N-3*X;T'k'X;P!2C23-NX;r'X;EB<=2NX;T'X;U2;T->B*X;S';‚;g302>X;Q';H:99dJ';S#B)32<(-024;)@-62<1*<C*B.@2;
>26)01>-M*3;D*>;1!>22<4)@230)*3-B;2B2+1>*3;@)+>*0+*#?';gB1>-@)+>*0+*#?;BVDHsJX;o9o<cn';
Y-4*3;1>-30D*>@
$;Y-42>@-+!2>X;Q';HcmmsJ';P!>22<4)@230)*3-B;>2+*301>.+M*3;D>*@;>-34*@;#>*I2+M*30$;*>)231-M*3-B;-B)63@231;C)-;Y-4*3;1>-30D*>@0';9#-/*)$1/(21('8#FD$;c:c•co8';
w-C2B21;1>-30D*>@$
;S*>N-3*X;T'k'X;x*3)+X;S'X;EB<=2NX;T'X;T->-N*X;x'Q'X;U2;T->B*X;S'X;P!2C23-NX;r';‚;g302>X;Q';H:99sJ';[;@.BM>20*B.M*3;-##>*-+!;1*;*>)231-M*3;-00)63@231;)3;oU;2B2+1>*3;@)+>*0+*#?;*D;0)36B2;#->M+B20';x;
S1>.+1;=)*B;BEWHoJX;oic<m:';
r*B->;O*.>)2>;1>-30D*>@$;
=-/2>X;P'S';‚;T!236X;Y'v';Hcmm8J';[;@*42B<(-024;-##>*-+!;D*>;4212>@)3)36;*>)231-M*30;*D;()*B*6)+-B;@-+>*@*B2+.B20;)@-624;(?;+>?*2B2+1>*3;@)+>*0+*#?';x;S1>.+1;=)*B;BBWHcJX;c:9<o9';
S*€&->2$
;;SrKUEYX;KQ[uKTX;^QKrrX;EQ[FX;21+';L'>@70,#J7(#('C-'&'->#7J#%+*2-&'->#%-0#('$7-,>(9$67-;;
L'>@70,#>@%>#P9%-6X'#>@'#%+*2-&'->#8%(%&'>'(,#:B;#
!"#$%&'($
)&*'$%&'($
U)>2+1;-">)(.M*3;*D;E.B2>;-36B20;1*;1!2;
2,#2>)@231-B;)@-620;(?;+*@#->)0*3;
&)1!;:U;#>*I2+M*30;*D;-;B*&2><
>20*B.M*3X;>2D2>23+2;C*B.@2';;
I'J'('-$'#8(7O'$67-,;H#>*I2+M*30;*D;-;>2D2>23+2;C*B.@2X;
4)0+>212;3.@(2>;*D;#>*I2+M*3;4)>2+M*30J;
L'>@70,#>@%>#P9%-6X'#>@'#%+*2-&'->#8%(%&'>'(,#:3;#
k32;*D;1!2;6B*(-B;>2G32@231;01>-126)20';
E62B@-3;:999X;gB1>-@)+>*0+*#?X;id$;::d<:os';