Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
The IJB-A Face Verification Challenge
Performance Report
Patrick Grother and Mei Ngan
Caution: This report quantifies face recognition performance using data supplied byexternal research and development organizations. Its results are derived from self-administered experiments on the fully public IJB-A dataset. As such the results canbe manipulated by various means that may not be operationally realistic. Therefore,end users of face recognition technology should prefer results from NIST’s ongoingsequestered testing campaigns, FRVT or FIVE, or on similar independent evaluationsof face recognition. Developers whose algorithms exhibit good performance here areencouraged to submit their algorithms into those sequestered tests programs.
This report is generated automatically. It will be updated as new algorithms areevaluated, and as new analyses are added. Automated notifications can be obtainedvia the mailing list. Correspondence should be directed to the authors via [email protected].
This report was last updated on April 28, 2017.
http://www.nist.gov/itl/iad/ig/facechallenges.cfmhttp://www.nist.gov/itl/iad/ig/frvt-home.cfmhttp://www.nist.gov/itl/iad/ig/five.cfmmailto:[email protected]?subject=subscribe
IJB-A 1:1 VERIFICATION 1
Figure 1: Three images of one subject in the IJB-A dataset. The entire dataset is available online. Many photos were taken by photo journalistsand, as such, are well exposed, well focused, and deemed suitable for public display. For face recognition, they nevertheless remain challengingdue to wide variations in pose, illumination, expression and occlusion.
1 Introduction
Three IARPA Janus Benchmark A challenges are described by Klare et al. in the paper Pushing the Frontiers ofUnconstrained Face Detection and Recognition[2]. The first of these, the IJB-A 1:1 challenge, quantifies performanceof face verification algorithms (“same person or not?”) on challenging photo-journalism images of the kind shownin Figure 1. They are considerably more difficult to recognize than the portraits mandated by facial recognitionstandards1.
IJB-A 1:1 is a “take-home” test in that it is based on fully public data. It follows the design of the LFW protocol inrequiring many pairs of samples to be compared in isolation2. This corresponds to recognition tasks like passportverification or forensic comparison where there is just a pair of samples and no central database or gallery.
The IJB-A 1:1 challenge departs from LFW as follows:
. Face selection: LFW contains faces that could be detected with the Viola-Jones face detection algorithm. Thislimits difficulty. IJB-A on the other hand, uses manually located and annotated faces.
. Landmarks: The IJB-A tests include landmark coordinates (eyes and nose) whereas LFW provides just rawimages, and aligned (funneled) images.
. Multi-image samples: LFW compared single images. IJB-A uses richer samples containing 1 ≤ K ≤ 202images, including frames from video sequences.
. More impostor pairs: IJB-A 1:1 uses many more impostor comparisons that genuines. In LFW, the ratio was 1which precluded computation of false match rates at usefully low values.
2 Metrics
This section describes the one-to-one verification accuracy metrics present in this report.
1 NIST maintains a challenge for such images based on the mugshots of NIST Special Database 32 (“MEDS”)[1]. This is intended as a steppingstone prior for developers prior to entering NIST’s ongoing fully sequestered FRVT verification test.
2IJB-A 1:1 does not cross-compare galleries and probesets; it has no concept of such. It does not attempt to measure both verification andidentification accuracy from the same similarity score matrix; it does not pin the prior probabilities of impostor vs. genuine pairs i.e. O(n2) vs.O(n).
NIST 1 Last updated: April 28, 2017
http://www.nist.gov/itl/iad/ig/facechallenges.cfmhttp://www.nist.gov/itl/iad/ig/facechallenges.cfmhttp://www.nist.gov/itl/iad/ig/frvt-home.cfm
IJB-A 1:1 VERIFICATION 2
2.1 Quantifying false acceptance
False acceptance is computed over N comparisons. Each comparison involves a pair of samples, one from each oftwo different individuals. Each comparison yields a single non-negative scalar similarity score. The false match rate(FMR) is defined as the proportion of scores that are at or above threshold T .
FMR(T ) = 1− 1N
N∑i=1
H(si − T ) (1)
where H is the Heaviside unit step, and si is the score from the i-th impostor comparison.
2.2 Quantifying false rejection
False rejection is computed over M comparisons. Each comparison involves a pair of samples. Both samples comefrom a single individual. Each comparison yields a single non-negative scalar similarity score. The false non-matchrate is defined as the proportion of scores that are below a threshold T .
FNMR(T ) =1
M
M∑i=1
H(si − T ) (2)
with si being the score from the i-th genuine comparison.
The common term true accept rate (TAR) is a synonym for the complement 1-FNMR. This document uses neitherTAR nor false accept rate (FAR), as these are reserved[3] for transactional error rates involving potentially severalattempts to use a biometric system. The terms here, FNMR and FMR, are matching error rates, befitting IJB-A.
2.3 Quantifying failure to enrol
The above definitions assume that each comparison produces a score. Indeed the IJB-A protocol requires a completeset of scores to be submitted. This is operationally atypical because: a) some algorithms electively refuse to enrolimagery that is too poor to process; b) face or landmark detection can fail; c) feature extraction fails; and d) soft-ware throws an exception. Together these outcomes are quantified by the failure-to-enrol rate (FTE), a term whosedefinition is overloaded in the literature, and in practice.
This report includes two statements of the failure to enrol rate (FTE).
. Empty templates: The proportion of all templates that are empty, where empty is defined as having size below32 bytes. This value is used heuristically to handle implementations that sometimes fail to produce a viabletemplate but nevertheless include a short header.
. Failed comparisons: The proportion of comparisons that give a non-zero return code, or a negative similarityscore.
The consequences of FTE operationally should depend on the application. A one-to-one access control systemwould reject the presentation, and allow limited re-submissions of new face samples. In negative identificationsystems, where subjects make an implicit claim not to be present in a database (e.g. deportees), the correct actionwould be to investigate the sample for evidence of evasion or tampering. To effect fair comparison of algorithms,failed comparisons must be accounted for. This is done here by setting scores to zero to simulate rejection. Thiscorresponds operationally to a (fortuitously) correct rejection of an impostor, and an incorrect rejection of a genuineuser. Note that in the case of negative identification, scores should be set to high values to simulate the triggeringof a human investigation. This would correctly flag enrolled subjects, and incorrectly flag the non-enrolled. Mostresearch reports set scores to zero, as is done here.
NIST 2 Last updated: April 28, 2017
IJB-A 1:1 VERIFICATION 3
This issue presents a gaming opportunity: An algorithm developer can code a strategy for handling low qualityimages if he knows or assumes the number of impostor comparisons in a test will be larger (or smaller) than thenumber of genuines. In such cases, his implementation would speculatively and preferentially guess a low (high)score in cases where feature extraction fails, or when image quality is poor. While the IJB-A challenge is wide opento such gaming, there are several mitigations to such strategies, particularly in sequestered tests. Operationally, anoff-the-shelf implementation will not know the prior probability of an impostor.
2.4 Score-range normalization
Some graphs in this report include similarity scores normalized via the probability integral transform. If the empir-ical cumulative distribution of the comparison scores is F , then the new variable
s† = F (s) (3)
will be uniformly distributed on [0,1] (absent tied scores). This transformation is applied later to compare theresponse of different algorithms to certain covariates.
If the transform is applied to just the genuine scores, then for a threshold, T, the quantity F(T) is the false non-matchrate, FNMR(T), i.e. the fraction of genuine scores below threshold. Likewise a function G(T) computed over just theimpostor scores gives FMR(T) = 1 - G(T).
The function F, computed from genuine scores, can also be applied to impostor scores.
3 Results
3.1 Comparing accuracy
The graphs that follow include results for several classes of algorithms that are differentiated by their developmentdate, and use of landmarks and training data - see Table 1. This latter issue is nuanced and yet critical to under-standing how and whether algorithms can be compared. Historically commercial algorithms have been providedand used in an entirely off-the-shelf manner - the representation is fixed and the user in no way adapts (trains)the algorithm to his native data. The academic community, meanwhile, almost always isolates some portion of thedata for the express purpose of adapting the algorithm. The result is a refined set of parameters, or explicit data“models” (most prosaically, a PCA basis set). The academic community, ignoring marketplace practice, has notedthat recognition accuracy is improved by training, and training is improved through detailed exploitation of largetraining sets. Why then do commercial implementations not roll-out training facilities within their commercial offthe shelf products. The answer partly rests on the observation that succesful training and adaptation is a fine artthat, empirically, cannot be canned in a simple function call.
That said, one particular kind of training is possible operationally: Gallery training occurs after templates havebeen enrolled into a gallery. This is discussed in the accompanying 1:N report. The issue is moot here because 1:1verification exists in the IJB-A challenge without any gallery ever being constructed.
3.2 Performance gains through time
Figure 10 shows DETs by organization. The DETs generally march down and to the left.
NIST 3 Last updated: April 28, 2017
IJB-A 1:1 VERIFICATION 4
Algorithm Developmentthru
Training data Role of provided IJB-A landmarks
BENCH-MK1 2010 External training data onlyANON-2013 2013 External training data only Algorithm was provided only with
image cropped from bounding boxMSU-071715 2015 External training data and IJB-A
training splitsAlgorithm was provided only withfull IJBA-specified landmarks
JANUS* 2015-09 External training data and IJB-Atraining splits
Algorithm was provided only withfull IJBA-specified landmarks
RankOne-011816 2016-01 External training data Developer asserts IJBA boundingboxes and landmark points were notused.
Table 1: Context of use: Comparison of the algorithms should be conducted in the context of variations in when they were developed, withwhat training data and on whether the ran in fully automated mode or were assisted by the provision of geometric information. Note thatthe ANON-2013 algorithm was developed before the IJB-A challenge was assembled and was provided to NIST without the expectation that itwould be run on images of this type.
FTE EMPTY TEMPLATES FTE FAILED COMPARISONS
BENCH−MK1
JanusB−071315
JanusB−092015
JanusC−071515
JanusC−090815
JanusD−071715
MSU−072115
NUS−Panasonic−032917
VCOG−021317
COTS−04
COTS−08
RankOne−011816
RankOne−091015
0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3Algorithm
Failu
re to
Ext
ract
Rat
e
Failure to enrol, updated 2017−04−27 18:13:14
Figure 2: Failure to enrol: Per the discussion in section 2.3, the chart quantifies failure to enrol rate (FTE) in two ways: as the proportion ofempty templates; and as the proportion of failed comparisons. These two quantities are not independent since empty templates should not yieldvalid comparison scores.
NIST 4 Last updated: April 28, 2017
IJB-A 1:1 VERIFICATION 5
0.041
0.155
0.178
0.227
0.265
0.275
0.303
0.421
0.448
0.543
0.561
0.741
0.803
NUS−Panasonic−032917
VCOG−021317
JanusB−092015
JanusD−071715
JanusB−071315
MSU−072115
JanusC−090815
RankOne−011816
BENCH−MK1
JanusC−071515
RankOne−091015
COTS−08
COTS−04
0.0 0.2 0.4 0.6 0.8FNMR@FMR=0.01
Figure 3: Leaderboard: The chart shows the false non-match rate (FNMR) at a false match rate (FMR) of 0.01, as a headline comparativeaccuracy statement. This FMR value is higher than that typically targeted in operational face verification settings, but is chosen here given thelimited number of subjects present in the IJB-A 1:1 dataset. The effects of failure-to-enrol are included in these accuracy numbers. The full errortradeoff characteristics of Figure 4 give accuracy estimates at a broader range of FMR values.
NIST 5 Last updated: April 28, 2017
IJB-A 1:1 VERIFICATION 6
0.04
1
0.15
5
0.17
8
0.22
7
0.26
5
0.27
5
0.30
3
0.42
1
0.44
8
0.54
3
0.56
1
0.74
1
0.80
3
NU
S−
Pan
ason
ic−
0329
17
VC
OG
−02
1317
Janu
sB−
0920
15
Janu
sD−
0717
15
Janu
sB−
0713
15
MS
U−
0721
15
Janu
sC−
0908
15
Ran
kOne
−01
1816
BE
NC
H−
MK
1
Janu
sC−
0715
15
Ran
kOne
−09
1015
CO
TS
−08
CO
TS
−04
0.0
0.2
0.4
0.6
0.8
FN
MR
@F
MR
=0.
01
0.25
0.50
0.75
1e−
041e
−03
1e−
021e
−01
FM
R
FNMRA
lgor
ithm
BE
NC
H−
MK
1
CO
TS
−04
CO
TS
−08
Janu
sB−
0713
15
Janu
sB−
0920
15
Janu
sC−
0715
15
Janu
sC−
0908
15
Janu
sD−
0717
15
MS
U−
0721
15
NU
S−
Pan
ason
ic−
0329
17
Ran
kOne
−01
1816
Ran
kOne
−09
1015
VC
OG
−02
1317
DE
Ts
for
IJB
−A
1:1
cha
lleng
e, u
pdat
ed 2
017−
04−
27 1
8:08
:25
Figu
re4:
DET
s:A
tlef
t,ar
efu
ller
ror
trad
eoff
char
acte
rist
ics
(DET
s)fo
rth
eIJ
B-A
algo
rith
ms.
Atr
ight
isa
sum
mar
yst
atis
tic
corr
espo
ndin
gto
ave
rtic
alsl
ice
from
the
DET
s.T
his
isin
clud
edto
ease
look
up.W
hile
itim
plie
son
era
nkin
gof
the
algo
rith
ms,
itis
nota
ble
that
som
eD
ETs
cros
ssu
chth
atfa
lse
reje
ctio
nac
cura
cyde
pend
she
avily
onth
eFM
Rop
erat
ing
poin
t.A
lgor
ithm
deve
lope
rsca
nsh
ape
the
DET
char
acte
rist
ics
acco
rdin
gto
akn
own
use-
case
.Muc
hac
adem
icre
sear
chop
tim
izes
atFM
Rva
lues
high
erof
ten
than
isop
erat
iona
llyus
eful
.Som
eal
gori
thm
sex
hibi
tflat
ter
DET
s.
NIST 6 Last updated: April 28, 2017
IJB-A 1:1 VERIFICATION 7
BE
NC
H−
MK
1C
OT
S−
04C
OT
S−
08Ja
nusB
−07
1315
Janu
sB−
0920
15Ja
nusC
−07
1515
Janu
sC−
0908
15Ja
nusD
−07
1715
MS
U−
0721
15N
US
−P
anas
onic
−03
2917
Ran
kOne
−01
1816
Ran
kOne
−09
1015
VC
OG
−02
1317
0.25
0.50
0.75
0.25
0.50
0.75
0.25
0.50
0.75
0.25
0.50
0.75 1
e−04
1e−
031e
−02
1e−
01F
MR
FNMR
split
1 2 3 4 5 6 7 8 9 10
DE
Ts
for
IJB
−A
1:1
cha
lleng
e, u
pdat
ed 2
017−
04−
27 1
8:08
:30
Figu
re5:
DET
sby
spli
t:D
ETs,
asbe
fore
,but
wit
hin
divi
dual
trac
esfo
rea
chof
the
ten
split
spr
esen
tin
the
IJB-
Ase
t.O
ff-t
he-s
helf
reco
gnit
ion
engi
nes
typi
cally
give
tigh
tly
clus
tere
dD
ETs
-thi
sis
anat
trac
tive
trai
t-se
vera
loth
ers
give
vari
able
DET
s.St
atis
tica
lsig
nific
ance
islo
stat
the
left
side
ofea
chpa
nel(
atlo
wFM
R).
NIST 7 Last updated: April 28, 2017
IJB-A 1:1 VERIFICATION 8
●●
●
●
●
●
●
● ●
●
●
BE
NC
H−
MK
1
CO
TS
−04
CO
TS
−08
MS
U−
0721
15
Ran
kOne
−09
1015
Ran
kOne
−01
1816
Janu
sB−
0713
15
Janu
sB−
0920
15
Janu
sC−
0715
15
Janu
sC−
0908
15
Janu
sD−
0717
15
VC
OG
−02
1317
NU
S−
Pan
ason
ic−
0329
17
0.04
0.05
0.06
0.07
0.08
0.09
0.10
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1e+
033e
+03
5e+
031e
+04
3e+
045e
+04
1e+
05Te
mpl
ate
Siz
e (b
ytes
)
FNMR @ FMR = 0.01
Alg
orith
m
● ● ● ● ● ● ● ● ● ● ● ● ●
BE
NC
H−
MK
1
CO
TS
−04
CO
TS
−08
Janu
sB−
0713
15
Janu
sB−
0920
15
Janu
sC−
0715
15
Janu
sC−
0908
15
Janu
sD−
0717
15
MS
U−
0721
15
NU
S−
Pan
ason
ic−
0329
17
Ran
kOne
−01
1816
Ran
kOne
−09
1015
VC
OG
−02
1317
Asy
mm
etric
●FA
LSE
TR
UE
Acc
urac
y vs
. tem
plat
e si
ze tr
adeo
ff fo
r IJ
B−
A 1
:1 c
halle
nge,
upd
ated
201
7−04
−27
18:
09:3
8
Figu
re6:
Acc
urac
yvs
.te
mpl
ate
size
trad
espa
ce:
The
grap
hpl
ots
asu
mm
ary
FNM
Rva
lue
(atF
MR
=0.
01)
agai
nstt
hesi
zeof
asi
ngle
-im
age
tem
plat
e.W
hile
larg
ete
mpl
ates
noti
onal
lyof
fer
info
rmat
ion-
rich
repr
esen
tati
ons
and
impr
oved
accu
racy
,the
ym
ayal
sosu
ffer
curs
e-of
-dim
ensi
onal
ity
effe
cts.
Ther
eis
nocl
ear
indu
stry
-wid
etr
ades
pace
evid
ent.
An
asym
met
ric
algo
rith
mha
sdi
ffer
ente
nrol
men
tand
veri
ficat
ion
tem
plat
esi
zes.
NIST 8 Last updated: April 28, 2017
IJB-A 1:1 VERIFICATION 9
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●●
●
●●
●●
●
●
● ●
● ●
●
●●
●
●●
●●
●
●
● ●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
● ●
●
●
●
●
●
●
●
●●
●
●
● ●
●
●
●
●
●●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●
●●
●
●●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●
●
● ●
●
●●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●●
● ●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●●
● ●
●
●
●● ● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●●
●
●●
●●
●
●●
●
●
●
●●
●
●●●
●
●●
●
●
●●
●
T(1
) =
905
76,
T(n
) ~
516
30n^
1.1
byte
s
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●
●
● ●
●
●●
●●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●● ● ●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●●
●
●
●
●●
●
●●
●
T(1
) =
160
00,
T(n
) ~
160
00n^
1.0
byte
s
●●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●●
●
●●
●●
●
● ●
●●
●
●●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
● ●●
●
●
●
● ●
●
●
●
●
●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●●
●
●
●
●
● ●
●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●
●
● ●●
●●
●
●
●
●
●
●●
●
●●
●
●●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●● ● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●●
●
●●
●
●
●
●
●●
●
●
●
●
●
●●●
●
●
●●
●●
●
●
T(1
) =
560
0,
T(n
) ~
419
7n^1
.0 b
ytes
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●
●
● ●
●
●●
●●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●● ● ●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●●
●
●
●
●●
●
●●
●
T(1
) =
128
000,
T
(n)
~ 1
2798
2n^1
.0 b
ytes
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●
●
● ●
●
●●
●●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●● ● ●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●●
●
●
●
●●
●
●●
●
T(1
) =
937
52,
T(n
) ~
937
39n^
1.0
byte
s
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●
●
● ●
●
●●
●●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●● ● ●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●●
●
●
●
●●
●
●●
●
T(1
) =
411
2,
T(n
) ~
410
7n^1
.0 b
ytes
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●
●
● ●
●
●●
●●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●● ● ●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●●
●
●
●
●●
●
●●
●
T(1
) =
414
4,
T(n
) ~
413
9n^1
.0 b
ytes
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●●
●●
●●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●●●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
T(1
) =
163
84,
T(n
) ~
163
84n^
0.0
byte
s
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●
●●
●
●
●
●●
●
●
● ●
●
●●
●●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●
●● ● ●
●
●
●
●
●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●●
●
●
●
●●
●
●●
●
T(1
) =
400
, T
(n)
~ 4
00n^
1.0
byte
s●●●●●●●●●
●●●
●
●
●
●●●●●●
●●
●
●●●●
●
●
●
●●●
●●
●
●
●●●●●
●●
●
●●●●
●●●
●
●
●●●●●
●
●
●
●●●●●
●●
●
●●●●●
●
●
●
●●●●●●●●
●●
●
●●●●●●●
●
●
●
●●●●●●●●●●●●●●
●●
●
●
●
●●●●●●●●●●●●●●
●
●
●
●●
●●
●
●
●●●●
●
●●
●
●●●●●●●
●●●●
●
●●●●●●
●
●
●
●●●●●
●●
●
●●●●●
●●
●
●●●●●●●●●
●●
●
●
●●●●●●●●●●
●●●
●
●
●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●●
●
●
●
●●●●●●●●
●
●
●
●●●●
●
●
●
●
●●●●●●
●
●
●
●●●●●●●●
●
●●
●
●●●●●●●●●
●
●
●
●
●●●●●●●●●●●●●●●●●
●●
●
●
●●●●
●
●
●
●●●● ●
●
●●●●●
●
●
●
●●●●
●●
●
●●●●●●
●
●●●●●●
●●
●
●
●
●●●
●●
●
●
●●●●●●●●●●●●●●●●●●●●●●
●
●●
●
●●●●●●●
●●
●●
●
●●●●●●●●●●●●●●●
●●
●●
●●
●●
●
●
●
●●●●●
●●
●●
●
●●●
●●
●
●
●●●●●●●
●●●
●
●
●●●●●●●
●●●
●
●
●
●●●●●●●●●●
●●
●●
●
●
●
●●●●●●●●●●
●
●
●●●●●●●●●●●●●●●●●
●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●●●●●
● ●
●
●●●●●●●●●●
●
●
●
●●
●●●
●●●●
●●
●
●
●●●●
●●●●
●●
●●
●
●
●
●●●
●●
●
●
●●●●
●
●
●
●●●●●
●●
●
●●●●●●●
●●
●●
●
●●●●●
●●
●
●●●●●●●
●●
●
●●●●●●●●●●●●●●●●
●●
●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●●● ●●●●●●●●●●●●●●●●
●●●
●●
●
●
●●●●
●
●
●
●●●●●●●
● ●
●
●●●●●●●●●●●●●
●
●
●
●
●●●●●
●●
●●●
●
●
●●●●●●
●●
●●
●
●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●
●●●●●●●●●●●●
●●
●
●
●
●●●●●●●●●
●
●
●
●●●●
●
●
●
●●●●●●● ●●●●●●●
● ●
●
●●● ●●●●
●
●●
●
●●●● ●●●●●●
●
●●●●● ●●●
●●
●●
●
●●●●●●
●●●
●
●●●
●
●
●
●
●●●
●●●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●●
●
●
●
●
●●●●●●●●●●●
●●
●●
●
●
●
●
●●●●●●●
●
●
●●●●●●●
●
●
●
●●●●●●●
●
●
●●●● ●
●
●●●
●●
●
●●●
●●●●●●
●
●
●
●
●●●●●●●●●●●●●●●●●●●●
●●
●●
●
●
●
●●●
●●●
●●●●●●
●
●●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●
●●
●
●
●
●
●●●●●●●●
●●
●
●●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●●●●●●
●●
●
●●●● ●
●
●●●●●
●
●
●
●
●●●●●●
●●● ●●●●●
●●
●
●
●●●●●
●●
●
●●●●●●
●
●
●
●
●●●●
●
●
●
●●●●●
●●
●
●
●
●●●●●●●●
●
●
●
●●●● ●
●
●●●●●●●●
●
●
●
●
●●●
●●●●●
●
●
●
●●●●●
●
●●
●●●●●●
●
●
●
●
●●●●●●●●●●●●●●
●●●
●
●
●●●● ●
●
●●●●●
●●
●●
●
●●●●●●●●
●
●
●
●●●● ●
●
●●●●●●●●●●●●●
●●●●●●
●●
●
●
●
●
●●●●●
●●
●
●●●●●●●●●●●
●
●
●
●
●●●●●●
●
●
●
●●●●●●●●●●●
●
●
●
●●●●●●●●
●
●●
●●●●●
●●●
●
●
●
●●●●●
●●
●
●●●●●●●●
●
●
●
●●●●●●●
●●
●●●●●●●●●●●●
●
●
●
●●●
●●
●
●●●●●●
●
●●●
●●●●●●●●
●●
●
●
●
●
●●●●
●
●●
●●●●
●●●
●●●●●●●●
●●●
●
●
●●●●●●●
●
●●●●●●●●
●●
●
●●●●●●●
●●●
●
●
●●●
●
●
●
●
●●●●●●●●
●
●
●
●●●●●●
●
●●●
●●
●
●●●●●●●●●●
●●
●
●●
●
●●●●●●
●●
●
●●●●
●
●
●
●●●●●●●
●
●
●●●●●●
●●
●
●●●●●●●
●●
●
●
●
●●●●●●
●●
●●
●
●
●
●
●●●●●●●●
●
●
●
●●●●
●
●
●
●●●
●●
●
●●●●●●
●
●
●
●●●●●●●
●●
●
●
●
●●●
●●●●●●●●●●
●●
●
●
●
●●●
●●●● ●●●●●●●●●●●●●●●●●●●●●●●●
●●
●●●
●
●
●●●●●●●
●●
●
●●●●●● ●
●
●●●●●●●●
●
●
●
●●●●●●
●●●
●
●●●●●●●●
●●
●
●
●
●●●●●●●●●●●●●●●
●●
●
●
●
●●●●
●●
●
●●
●
●●●●●●
●●●
●
●
●●●●● ●
●
●●●●
●
●
●
●●●●●
●
●
●
●●●●●●
●
●
●
●●●●●●●●
●
●
●●●
●●
●
●●●●●
●
●●
●
●●●●
●●
●
●●●●
●●
●
●●●●●●
●
●●●●
●
●
●
●●●●●●●
●
●
●
●●●●
●●
●
●
●●●●●●●
●●
●
●
●
●●●●●
●●
●
●
●
●
●●●●●
●● ●
●
●●●
●●
●●
●
●●
●
●●●●●●
●
●●●●●●●●●●●●●●●●●
●●●
●
●
●
●●●●●
●●●
●
●
●
●●●●●●
●●●●●
●
●
●
●●●●●●●●●
●●●●
●
●
●
●●●●●
●
●
●
●
●●●●
●●
●
●
●
●●●●●●●●●●
●
●
●
●
●●●●●●●●●●●●●●●●
●●
●
●
●
●●●●●
●
●
●
●
●●●●●●
●
●
●
●●●●
●●●
●
●
●
●●●●
●●
●
●
●
●●●●●●●●
●●
●
●
●●●●●●
●●
●
●●
●
●●●●●
●●
●●
●
●●●●●●●●●●●●●●●●●●●
●
●
●
●●●●●●●
●
●
●
●●●●●●●●●●●●
●
●
●
●●●●●●●
●●
●
●
●●●●
●●
●
●●●●●●
●
●
●●●●●●
●
●●
●
●●●●●●●●●
●
●
●
●
●●●●●
●
●
●
●●●●●●
●
●●●●●●
●●●
●
●
●
●●●●
●
●
●
●●●●●●●●●●●
●
●
●
●●●●●●●
●●
●
●●●●●
●
●●
●
●●●●●●●●●●●●●●●●●●●●
●●●●●●
●
●
●
●●●●●●
●
●
●
●
●●●● ●
●
●●●●
●
●
●
●●●●●●
●●
●
●●●●
●●
●
●
●
●●●
●●
●
●●●
●●
●
●
●●●●●●
●●
●
●
●
●●●
●●
●
●●●●●●
●
●
●
●●●●
●
●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●●●●●●
●
●
●
●●●●●● ●
●
●●●●●●
●
●●●●●●●●●●●
●●
●
●●●●●●●●●●●●
●●
●●●●
●
●
●●●●●●●
●
●
●
●
●●●●●●●●●●
●
●
●
●●●●●●●●●●●
●●
●
●●●●●●●●●●●●●
●●
●●
●●
●
●
●
●●●●
●●
●
●
●●●●●●●●●●●●
●
●
●
●●●●●●●●●●●
●
●
●●●●●●●●●●
●
●
●
●●●●●
●●
●●
●
●●●●●● ●
●
●●●●●●
●●
●●
●●
●
●●●
●●
●
●
●
●●●●●●●●
●
●
●
●●●●●●●
●●●
●●
●
●●
●
●●●●●●●●●●
●
●
●●●
●●
●
●
●
●●●●●●● ●
●
●●●●●
●●
●
●●●●●●●●
●
●
●
●●●●●●●
●●
●
●●●●●●●●●●
●
●
●
●●●●
●●
●
●●●
●●●● ●
●
●●●●●●●●
●
●
●
●●●
●
●
●
●
●●● ●●●●
●
●
●
●●●●
●
●●
●●●●
●
●
●
●●●●●●●●
●
●
●
●●●●●●●●●●●●●●●●●●
●●●●●
●●
●
●
●
●●●●●●●●
●●
●●
●●●●
●●
●●●●●●●●●●●
●
●
●
●●●●● ●
●
●●●●●●●●
●
●
●
●●●●●●●
●●
●
●
●●●●●●●
●●
●
●
●●●●
●
●
●
●●
●●
●
●
●●●●●●●
●
●
●
●●●
●●
●
●●●● ●
●
●●●●●● ●
●
●●●●●
●●●
●
●
●●●●● ●
●
●●●
●
●●●●●●
●●
●
●●●●●●●●●
●
●
●
●●●●●●●●●●●●●●
●
●
●
●●●●●●●●●●●●
●●
●
●
●
●●
●●
●●
●●●●
●
●
●
●●●●●●●●●●●●●
●●
●
●
●●●●●●
●
●
●
●●●●
●
●
●
●
●●●●●●●●
●●
●
●
●
●●●●●●
●
●
●
●
●●●●●
●
●
●
●●●●●●●●
●
●
●
●
●●●●●●●●●●●●●
●
●
●
●
●●●●●
●●
●
●●●●●●●●●●
●
●
●
●●●●●●
●●
●●●●●●●●
●
●
●
●●●●
●
●
●
●
●●●●●
●●
●
●●
●●
●●
●
●●●●●●
●
●●●●
●
●
●
●●●●●●
●
●
●
●●●
●●
●●
●
●●
●●
●
●
●●●
●●●
●●●●●●●●●●●●
●●
●
●
●
●
●●●● ●
●
●●● ●●●●
●●●
● ●
●
●●●●
●●
●
●●●●●●
●
●
●
●●●●●●
●
●
●
●●●●●●●●●
●
●
●
●●●
●●
●
●
●
●●●●●
●●
●
●●●●●●●●●●●●●●●●●
●●
●●
●●●
●
●
●●●●●●●●●
●
●
●
●●●●●●●●●●
●●
●●
●
●
●
●●● ●●●●●●
●● ●
●
●●●●
●
●
●
●●●●●
●
●
●
●
●●●●
●
●●
●
●●●●
●●●
●
●
●●●●●
●
●
●
●●● ●●●
●●
●●
●
●
●
●●●●●
●
●
●
●
●●●●●
●
●
●
●●●●●●●
●
●
●
●
●●●●●●●●
●
●●
●
●●●
●●●
●
●
●
●●●●
●
●
●
●
●●●●●
●
●
●
●
●●●●●●●●●●●●●●●●●●●
●
●
●
●
●●●●●●●
●●
●
●
●●●●●●
●
●
●
●●●●
●
●●
●
●●●●●●●
●
●
●
●●● ●●●●●●●●●●●●
●●●
●
●
●
●
●●●●●●
●
●
●●●●●●
●
●●
●
●●●●●
●●
●
●
●
●
●●●●●
● ●
●
●●●
●
●
●
●
●●●●●●●
●●
● ●
●
●●●●●●●●●●●
●
●
●●●●
●
●●
●●●●
●●
●
●
●
●●●●
●
●
●
●●●●●●●●●●●●●●●●●●●●●●●●●●
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
●
●
●
●●●●●
●●
●
●
●
●●●●●●●●●
●●●
●
●
●●●●●
●●
●
●
●
●●●
●●
●●
● ●
●
●●●● ●
●
●●●●●●●
●
●
●
●
●●
●●
●
●
●●●●●●●
●
●
●●●●●●●●●●●●●
●●●
●
●
●●●●●●●●
●●
●
●
●
●●●●●
●●
●
●●●●●●●●●
●
●
●●●●●
●●
●●
●
●●●●●●●
●
●
●
●●●●
●
●
●
●●●●●●●●
●
●
●●
●●
●
●
●●●● ●
●
●●●●●
●●
●
●●●●●
●
●
●
●
●●●●●●●
●●
●
●
●
●●
●●●
●●●●●●
●●
●
●●●
●●
●
●
●●●●●
●●
●
●
●
●●●●●●●
●
●●●●
●
●
●
●●●●●●●●●
●
●
●
●●●●
●●●
●
●●
●
●●●●●●●
●
●
●
●●●●●●
●●
●
●●●
●●
●
●
●
●
●●●
●●
●
●
●●●●●
●
●
●
●●●●●●
●
●
●
●●●
●
●
●
●
●●●●●
●
●
●
●●●●●
●●
●
●●
●●
●●
●●●●
●●●
●
●
●
●
●●●●●●
●
●●
●●●●●●●●●●
●
●
●
●
●●●●●
●●
●
●
●●●●●●●●●●●●●●●
●●●●
●
●
●
●●●●●●●●●●●●●●
●●
●
●
●
●●●●●●●
●●
●
●●●●●●●●●●●
●
●
●●●
●●
●
●●●●●●
●
●
●
●
●●●●●
●●
●●
●●
●
●●●●●●●
●●●
●
●
T(1
) =
245
76,
T(n
) ~
243
12n^
0.4
byte
s●
●
●
●
●●
●●
●
●
●
●●
●
●
●
●
●
●
●●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●●
●
● ●
●
●
●●
●
●●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
● ●
●
●
●
●
● ●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
● ●
●
●
●
●
●
●
●
●
●
●●●
●
●
●●
●
●
●
●
●
●● ●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●●
● ●
●
●
●●
●
●
● ●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●●
●●
●
●●
●
●
●●
●
●
●
●
●●
●
●
●●
●
●
●●●
●
●●
●
●
●●
●
T(1
) =
136
, T
(n)
~ 8
2n^1
.0 b
ytes
●●
●
●
●●
●
●
●
●
●
●●
●
●
●●
●
●●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
● ●●
●
●●
●
●
●
● ●
● ●
●
●
●
●
● ●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
● ●●
●
●
●
● ●●
●
●
●
●●
●
●●
●
●
●
●
●
●
●
●
●
●● ●
●
●
●
●
●●
●
●
● ●
●
●
●
●
●
●●
●
● ●
●●
●
●●
●
●
●
● ●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●●●
●●
● ●
●
●
●
●
●
● ●●
●
●
●●
●
●●
●
●
●
●
●●
●
●●
●
●
●
●
●●
●
●
●
●
●●
●
●●
●
●●
●●
●
●
●
●
●
●
●●
●●
●●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●
●●
●
●
●
●● ●
●
●
●
●
●
●●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●●
●
●
●
●●
●
●●●
●
●●
●●
●
●
●
●
T(1
) =
128
, T
(n)
~ 6
7n^1
.0 b
ytes
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●●
●●
●●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●●●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●
●●●
●●
●●
●●
●●
T(1
) =
820
8,
T(n
) ~
820
8n^0
.0 b
ytes
BE
NC
H−
MK
1C
OT
S−
04C
OT
S−
08Ja
nusB
−07
1315
Janu
sB−
0920
15Ja
nusC
−07
1515
Janu
sC−
0908
15Ja
nusD
−07
1715
MS
U−
0721
15N
US
−P
anas
onic
−03
2917
Ran
kOne
−01
1816
Ran
kOne
−09
1015
VC
OG
−02
1317
0.0e
+00
4.0e
+06
8.0e
+06
1.2e
+07
0
5000
00
1000
000
1500
000
2000
000
0e+
00
2e+
05
4e+
05
6e+
05
0.0e
+00
5.0e
+06
1.0e
+07
1.5e