large-scale quantitative analysis of painting arts (cabdyn, university of oxford, 2013)
TRANSCRIPT
Large-Scale Quantitative Analysis of Painting Arts
, Dongwoo Kim2, Seung-Woo Son
3, Alice Oh
2 , Hawoong
Jeong1,4
1 Department of Physics, KAIST, Republic of Korea2 Department of Computer Science, KAIST, Republic of Korea
3 Department of applied physics, Hanyang University, Republic of Korea4 Institute for the BioCentury, KAIST, Republic of Korea
Daniel Kim 1
PaintingStyle
Introduc-tion
Sum-mary
Color Spec-trum
Color Palette
CountingFre-
quency
FractalAnalysis
SurfaceAnalysis
Colorful Dots
Colorful Lines
Colorful Sur-faces
Stylometry
Litera-ture
Music
Period # imagesMedieval 331Early Renaissance 995Northern Renais-sance
1,047
High Renaissance 677Mannerism 911Baroque 3287Rococo 360Neoclassicism 163Romanticism 663Realism 146Total 8798
Data set 1: Web Gallery of Art (2009)
* Source: http://www.wga.hu/
Note * Title: Schloss Wilhelmshöhe with the Habichtswald c. 1800 Oil on canvas Neue Galerie, Kassel * Artist: German painter, Johann Erdmann Hummel (11 September 1769, Kassel — 26 Oc-tober 1852, Berlin)
P(r
)10-2
10-3
10-4
10-5
10-6
100 101 102 103 104 105
Rank: r
Null model
100 101 102 103 104 105 106
Rank: r
10-2
10-4
10-6
10-8
10-10
MedievalEarly Renaissance
Northern Renais-sanceHigh Renais-sanceMannerism
BaroqueRococo
NeoclassicismRomanticism
Realismphoto
P(r
)
PaintingStyle
Introduc-tion
Sum-mary
Color Usage
Color Palette
Count-ingFre-quency
FractalAnalysis
SurfaceAnalysis
Measuring“fractal dimen-sion“in RGB color space
)/(log
)(loglim)(
10
NSdbox
Blue
Red
Green
Measuring“fractal dimen-sion“in RGB color space
Blue
Red
Green
Th
e n
um
ber
of
non
-em
pty
boxes
Box side length
106
105
104
103
102
100 101 102
Box-c
ou
nti
ng
di-
men
sio
n
3
2.8
2.6
2.4
2.2
2Medieval
Early Renaissance
Northern Renais-
sance
High Renaissance
Mannerism
Baroque
Rococo
Neoclassicism
Romanti-cism
Real-ism
Period
Art historical considera-tion
In the medieval age …
1. Specific rare pigments were pre-ferred.
2. There was no physical mixing tech-nique.
Since the Renaissance…
1. The oil colors and the color mixing technique.
2. New painting techniques such as Chiaroscuro, Sfumato, Cangiante, …
PaintingStyle
Introduc-tion
Sum-mary
Color Usage
Color Palette
CountingFre-
quency
FractalAnalysis
SurfaceAnalysis
Two Painting Styles: Chiaroscuro & Sfu-mato
Painting Style 1: Chiaroscuro
↓Light ↓
Dark
Chiaroscuro ↓
Light ↓Dark
Brightness difference• Emphasis• Contrast• Perspective
Measuring“surface rough-ness"
Rough-ness ex-ponent
Brightness
Y-axis
2h(x)r)h(xG(r)
2r~
255
0
200400600
800
0
200400
600800
1000X-axis
0Note * Title: “St John the Evangelist Drinking from the Poi-soned Cup” * Artist: Italian painter Taddeo Gaddi (1348-1353)
r
104
103G
(r)
102
100
101
102
103
G(r)~ r2×0.28
Brightness
Y-axis
255
0
200400600
800
0
200400
600800
1000X-axis
0Note * Title: “St John the Evangelist Drinking from the Poi-soned Cup” * Artist: Italian painter Taddeo Gaddi (1348-1353)
Measuring“surface rough-ness"
0.40
0.35
0.309
0.25
0.20
〈α〉
Medieval
Early Renaissance
High Renaissance
Mannerism
Baroque
Rococo
Neoclassicism
Romanti-cism
Real-ism
Period
Northern Renais-
sance
Two examples of chiaroscuro
• Jackson Pollock
• Louis Wain
Jackson Pollock (1912-1956)
Note* Title: “Number 20, 1948, 1948”* Artist: American painter, Jackson Pollock (1912-1956)
Brightness
Y-axis
Brightness
X-axisY-axis
r
104
103
102
100 101 102 103
2α ~0.28 G
(r)
r100 101 102 103
101
102
103
104
105
G(r
)
2α ~0.008
255
0
400
800
0
400800
0
X-axis400
800
0
255
0
200
600
Louis Wain (1860-1939)
α0.30
0.25
0.20
0.15
0.10
0.050
Painting Style 2: Sfumato
Shading around eyes
Sfumato
Measuring“image en-tropy"
Brightness
Y-axis
255
0
200400600
800
0
200400
600800
1000X-axis
0Note * Title: “St John the Evangelist Drinking from the Poi-soned Cup” * Artist: Italian painter Taddeo Gaddi (1348-1353)
x )x(m
)x(plog)x(pH
p(x)=h(x)/S, m(x)=1+σ2(x)
y y
yhyhx
2
2 )(9
1)(
9
1)(
<
×104
Examiningcharacteristics of contracted im-ages
Radius ∝ color usage ~ mass
Red
Gre
en
4080
0
120
60
120
20
80
140
0
BlueCenter of mass
← Fixed point (FP)
↑ Shuffled Fixed point (SFP)
Shuffled image
↓
Blue
Red
Gre
en
4080
0
120
60
120
20
80
140
0
Medieval
Early Renaissance
High Renaissance
Mannerism
Baroque
Rococo
Neoclassicism
Romanti-cism
Real-ism
Period
Northern Renais-
sance
6
10
14
18
22
Submitted work
Ongoing work
Data set 2: Web Gallery of Art (2012)
Date # im-ages
1051-1100 311101-1150 221151-1200 201201-1250 351251-1300 1141301-1350 8041351-1400 2971401-1450 13281451-1500 31801501-1550 37351551-1600 19271601-1650 37531651-1700 22281701-1750 13421751-1800 9261801-1850 11451851-1900 1893Total 22780
* Source: http://www.wga.hu/
Data set 2: Web Gallery of Art (2012)
Date# im-ages
genre
histori-cal
inte-rior
land-scape
mythologi-cal
other
por-trait
reli-gious
still-life
study
1051-1100
31 0 0 0 0 0 0 0 31 0 0
1101-1150
22 0 0 0 0 0 2 0 20 0 0
1151-1200
20 0 0 0 0 0 1 0 19 0 0
1201-1250
35 0 0 0 0 0 2 0 33 0 0
1251-1300
114 0 0 1 0 0 0 0 113 0 0
1301-1350
804 0 1 15 4 22 7 3 752 0 0
1351-1400
297 0 3 8 0 2 1 4 278 0 1
1401-1450
1328 0 10 8 2 15 23 83 1187 0 0
1451-1500
3180 9 58 49 14 161 38 257 2590 3 1
1501-1550
3735 103 88 28 75 392 74 655 2313 6 1
1551-1600
1927 77 47 43 70 271 27 319 1029 42 2
1601-1650
3753 360 84 68 514 494 35 788 1011 379 20
1651-1700
2228 566 52 87 485 177 24 235 328 271 3
1701-1750
1342 243 41 15 279 203 10 240 216 92 3
1751-1800
926 141 42 20 241 61 18 282 89 28 4
1801-1850
1145 102 133 18 289 81 121 317 59 18 7
1851-1900
1893 323 33 16 757 50 156 407 27 105 19
Total 22780 1924 592 376 2730 1929 539 3590 10095 944 61
* Source: http://www.wga.hu/
Paintings are classified!
Data set 3: Google art project (2013)
* Source: http://www.google.com/culturalinstitute/project/art-project
-2550-2535-2520-2504-2489-2474-2459-2444-2428-2413-2398-2383-2367-2352-2337-2322-2307-2291-2276-2261-2246-2231-2215-2200-2185-2170-2155-2139-2124-2109-2094-2078-2063-2048-2033-2018-2002-1987-1972-1957-1942-1926-1911-1896-1881-1866-1850-1835-1820-1805-1790-1774-1759-1744-1729-1713-1698-1683-1668-1653-1637-1622-1607-1592-1577-1561-1546-1531-1516-1501-1485-1470-1455-1440-1424-1409-1394-1379-1364-1348-1333-1318-1303-1288-1272-1257-1242-1227-1212-1196-1181-1166-1151-1135-1120-1105-1090-1075-1059-1044-1029-1014-999-983-968-953-938-923-907-892-877-862-846-831-816-801-786-770-755-740-725-710-694-679-664-649-634-618-603-588-573-557-542-527-512-497-481-466-451-436-421-405-390-375-360-345-329-314-299-284-269-253-238-223-208-192-177-162-147-132-116-101-86-71-56-40-25-1052036516681971121271421571731882032182332492642792943093253403553703864014164314464624774925075225385535685835986146296446596756907057207357517667817968118278428578728879039189339489649799941009102410401055107010851100111611311146116111761192120712221237125312681283129813131329134413591374138914051420143514501465148114961511152615411557157215871602161816331648166316781694170917241739175417701785180018151830184618611876189119071922193719521967198319981
10
100
1000
10000
Year
# im
ag
es
Covering over 4000 years!
Data set 3: Google art project (2013)
979
994
1009
1024
1040
1055
1070
1085
1100
1116
1131
1146
1161
1176
1192
1207
1222
1237
1253
1268
1283
1298
1313
1329
1344
1359
1374
1389
1405
1420
1435
1450
1465
1481
1496
1511
1526
1541
1557
1572
1587
1602
1618
1633
1648
1663
1678
1694
1709
1724
1739
1754
1770
1785
1800
1815
1830
1846
1861
1876
1891
1907
1922
1937
1952
1967
1983
1998
1
10
100
1000
10000
Year
# im
ag
es
32808 paintings !
* Source: http://www.google.com/culturalinstitute/project/art-project
Data set 3: Google art project (2013)
979
994
1009
1024
1040
1055
1070
1085
1100
1116
1131
1146
1161
1176
1192
1207
1222
1237
1253
1268
1283
1298
1313
1329
1344
1359
1374
1389
1405
1420
1435
1450
1465
1481
1496
1511
1526
1541
1557
1572
1587
1602
1618
1633
1648
1663
1678
1694
1709
1724
1739
1754
1770
1785
1800
1815
1830
1846
1861
1876
1891
1907
1922
1937
1952
1967
1983
1998
1
10
100
1000
10000
Year
# im
ag
es
Dataset 2 + Dataset 3→ Over 55,000 paintings !
* Source: http://www.google.com/culturalinstitute/project/art-project
Data set 3: Google art project (2013)
Data set 3: Google art project (2013)
• Roughness exponents of “Web Gallery of Art images” (α vs. Year)
• Radius of gyration in RGB color space
What we are studying these days…
• If we iteratively shuffle an image until being a noisy image, how much time is required for each image?
• How many persons are appeared in paint-ing arts?
• How can we classify paintings by artists?
• Time evolution of roughness exponent in a video database.
Future work
Informaion Measure Represen-tation
Color types Fractal analysisColor palette diversity
Color types+ Color usage
Radius of gyrationColor palette diversity
Color types + Color usage+ Spatial correlation
Rescaling analysis Orderness
Spatial correlation Surface roughnessOrderness, painting style
Local spatial correla-tion Image entropy painting style
Sum-mary
Sum-mary• Massive painting databases are analyzed.
• Paintings and photos can be statistically dis-tinguished by rank-ordered color usage distri-bution shape.
• The changes of fractal dimension with the times can be historically interpreted as color palette expansion.
• Increasing trend of brightness surface rough-ness can be described as the art historical ren-ovation of painting technique and the diversifi-cation of painting genre.
• Jackson Pollock’s drip paintings are close to random images compared to European paint-ings based on roughness exponent values
Thank you!