Spaces:
Build error
Build error
File size: 65,509 Bytes
4dbe5d1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 |
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Fri Jun 19 09:40:58 2020
Image enhancement functions
@author: Vasileios Vonikakis (bbonik@gmail.com)
"""
import math
import imageio
import numpy as np
import matplotlib.pyplot as plt
from skimage.color import rgb2gray
from skimage import img_as_float
from skimage.exposure import rescale_intensity, adjust_gamma
plt.close('all')
#TODO: better memory management!!!! Too many copying of images.
#something like "inplace"?
def map_value(
value,
range_in=(0,1),
range_out=(0,1),
invert=False,
non_lin_convex=None,
non_lin_concave=None):
'''
---------------------------------------------------------------------------
Map a scalar value to an output range in a linear/non-linear way
---------------------------------------------------------------------------
Map scalar values to a particular range, in a linear or non-linear way.
This can be helpful for adjusting the range and nonlinear response of
parameters.
For more info on the non-linear functions check:
Vonikakis, V., Winkler, S. (2016). A center-surround framework for spatial
image processing. Proc. IS&T Human Vision & Electronic Imaging.
INPUTS
------
value: float
Input value to be mapped.
range_in: tuple (min,max)
Range of input value. The min and max values that the input value can
attain.
range_out: tuple (min,max)
Range of output value. The min and max values that the mapped input
value can attain.
invert: Bool
Invert or not the input value. If invert, then min->max and max->min.
non_lin_convex: None or float (0,inf)
If None, no non-linearity is applied. If float, then a convex
non-linearity is applied, which lowers the values, while not affecting
the min and max. non_lin_convex controls the steepness of the
non-linear mapping. Small values near zero, result in a steeper curve.
non_lin_concave: None or float (0,inf)
If None, no non-linearity is applied. If float, then a concave
non-linearity is applied, which increases the values, while not
affecting min and max. non_lin_concave controls the steepness of the
non-linear mapping. Small values near zero, result in a steeper curve.
OUTPUT
------
Mapped value
'''
# truncate value to within input range limits
if value > range_in[1]: value = range_in[1]
if value < range_in[0]: value = range_in[0]
# map values linearly to [0,1]
value = (value - range_in[0]) / (range_in[1] - range_in[0])
# invert values
if invert is True: value = 1 - value
# apply convex non-linearity
if non_lin_convex is not None:
value = (value * non_lin_convex) / (1 + non_lin_convex - value)
# apply concave non-linearity
if non_lin_concave is not None:
value = ((1 + non_lin_concave) * value) / (non_lin_concave + value)
# mapping value to the output range in a linear way
value = value * (range_out[1] - range_out[0]) + range_out[0]
return value
def get_membership_luts(
resolution=256,
lower_threshold=0.35,
upper_threshold=0.65,
verbose=False):
'''
---------------------------------------------------------------------------
Creates 3 paramteric traspezoid membership functions
---------------------------------------------------------------------------
The trapezoid functions are defined as piece-wise functions between the
0, lower_threshold, upper_threshold, 1. These trapezoid membership
functions can be used to filter out which parts of each exposure to be
used during exposure fusion. More details can be found in the following
paper:
Vonikakis, V., Bouzos, O. & Andreadis, I. (2011). Multi-Exposure Image
Fusion Based on Illumination Estimation, SIPA2011 (pp.135-142), Greece.
INPUTS
------
resolution: int
The size of the LUT (how many inputs).
lower_threshold: float in the range [0,1]
The position of the lower inflection point of the trapezoid functions.
It should be always lower compared to the upper_threshold.
upper_threshold: float in the range [0,1]
The position of the upper inflection point of the trapezoid functions.
It should be always higher compared to the lower_threshold.
verbose: boolean
Display outputs.
OUTPUT
------
lut_lower: float numpy array of size equal to resolution, values in [0,1]
The lower trepezoid membership function.
lut_mid: float numpy array of size equal to resolution, values in [0,1]
The middle trepezoid membership function.
lut_upper float numpy array of size equal to resolution, values in [0,1]
The upper trepezoid membership function.
'''
lut_lower = np.zeros(resolution, dtype='float')
lut_mid = np.zeros(resolution, dtype='float')
lut_upper = np.zeros(resolution, dtype='float')
for i in range(resolution):
i_float = i / (resolution - 1)
# lower trapezoid membership function
if i_float <= lower_threshold:
lut_lower[i] = i_float / lower_threshold
else:
lut_lower[i] = 1
# middle trapezoid membership function
if i_float <= lower_threshold:
lut_mid[i] = i_float / lower_threshold
elif i_float <= upper_threshold:
lut_mid[i] = 1
else:
lut_mid[i] = (1 - i_float) / (1 - upper_threshold)
# upper trapezoid membership function
if i_float <= upper_threshold:
lut_upper[i] = 1
else:
lut_upper[i] = (1 - i_float) / (1 - upper_threshold)
if verbose is True:
plt.figure()
plt.subplot(1,3,1)
plt.plot(lut_lower)
plt.title('Lower')
plt.grid(True)
plt.subplot(1,3,2)
plt.plot(lut_mid)
plt.title('Middle')
plt.grid(True)
plt.subplot(1,3,3)
plt.plot(lut_upper)
plt.title('Upper')
plt.grid(True)
plt.suptitle('Trapezoid membership functions')
plt.show()
return lut_lower, lut_mid, lut_upper
def get_sigmoid_lut(
resolution=256,
threshold=0.2,
non_linearirty=0.2,
verbose=False):
'''
---------------------------------------------------------------------------
Creates a paramteric sigmoid function and stores it in a LUT
---------------------------------------------------------------------------
The sigmoid function is defined as a piece-wise function of 2 inverse
non-linearities. This allows full control of the inflection point
(threshold) and the degree of 'sharpness' of each non-linearity. The
non-linear curves used here are described in the paper:
Vonikakis, V., Winkler, S. (2016). A center-surround framework for spatial
image processing. Proc. IS&T Human Vision & Electronic Imaging.
INPUTS
------
resolution: int
The size of the LUT (how many inputs).
threshold: float in the range [0,1]
The position of the inflection point of the sigmoid function (0.5 in
the mid_tonedle of the range).
non_linearirty: float in range (0, inf)
Controls the non-linearity of the curve before and after the inflection
point. It should not be 0. The smaller it is (asymptotically to 0) the
'sharper' the non-linearity. After ~5 it asymptotically approaches a
linerity.
verbose: boolean
Display outputs.
OUTPUT
------
lut: float numpy array of size equal to resolution
The output sigmoid lut.
'''
max_value = resolution - 1 # the maximum attainable value
thr = threshold * max_value # threshold in the range [0,resolution-1]
alpha = non_linearirty * max_value # controls non-linearity degree
beta = max_value - thr
if beta == 0: beta = 0.001
lut = np.zeros(resolution, dtype='float')
for i in range(resolution):
i_comp = i - thr # complement of i
# upper part of the piece-wise sigmoid function
if i >= thr:
lut[i] = (((((alpha + beta) * i_comp) / (alpha + i_comp)) *
(1 / (2 * beta))) + 0.5)
# lower part of the piece-wise sigmoid function
else:
lut[i] = (alpha * i) / (alpha - i_comp) * (1 / (2 * thr))
if verbose is True:
plt.figure()
plt.plot(lut)
plt.title('Sigmoid LUT | ' +
'thr=' + str(int(thr)) + ' (' + str(round(threshold, 3)) +
') | nonlin=' + str(int(alpha)) +
' (' + str(round(non_linearirty, 3)) + ')')
plt.grid(True)
plt.tight_layout()
plt.show()
return lut
def get_photometric_mask(
image,
smoothing=0.2,
grayscale_out=True,
verbose=False):
'''
---------------------------------------------------------------------------
Estimate the photometric mask of an image by using edge-aware blurring
---------------------------------------------------------------------------
Applies strong blurring while preserving the strong edges of the image in
order to avoid halo artifacts. Inspired by the paper:
Shaked, Doron & Keshet, Renato. (2004). "Robust Recursive Envelope
Operators for Fast Retinex."
INPUTS
------
image: numpy array (WxH or WxHxK of uint8 [0.255] or float [0,1])
Input image.
smoothing: float in the interval [0,1]
Value controlling the blur's strenght. 0 indicates no blur. Values
between 0-1 increase blurring strength while preserving edges. Values
above 1 approximate very strong gaussian blurring (large sigmas) where
no edges are preserved. Practically, values above 10 result into a
uniform image.
grayscale_out: logical
Whether or not the photometric mask is going to be grayscale or not.
If the input image is already grayscale (2D) then this parameter is
irrelevant.
verbose: boolean
Display outputs.
OUTPUT
------
image_ph_mask: numpy array of WxH or WxHxK of float [0,1]
Photometric mask of the input image.
'''
'''
Intuition about the threshold and non_linearirty values of the LUTs
threshold:
The larger it is, the stronger the blurring, the better the local
contrast but also more halo artifacts (less edge preservation).
non_linearirty:
The lower it is, the more it preserves the edges, but also has more
'bleeding' effects.
'''
# internal parameters
THR_A = smoothing
THR_B = 0.04 # ~10/255
NON_LIN = 0.12 # ~30/255
LUT_RES = 256
# get sigmoid LUTs
lut_a = get_sigmoid_lut(
resolution=LUT_RES,
threshold=THR_A,
non_linearirty=NON_LIN,
verbose=verbose
)
lut_a_max = len(lut_a) -1
lut_b = get_sigmoid_lut(
resolution=LUT_RES,
threshold=THR_B,
non_linearirty=NON_LIN,
verbose=verbose
)
lut_b_max = len(lut_b) -1
# dealing with different number of channels
if len(image.shape) == 3:
if grayscale_out is True:
image_ph_mask = rgb2gray(image.copy()) # [0,1] 2D
else:
image_ph_mask = img_as_float(image.copy()) # [0,1] 3D
elif len(image.shape) == 2:
image_ph_mask = img_as_float(image.copy()) # [0,1] 2D
else:
image_ph_mask = img_as_float(image.copy()) # [0,1] ?D
# if image is 2D, expand dimensions to 3D for code compatibility
# (filtering assumes a 3D image)
if len(image_ph_mask.shape) == 2:
image_ph_mask = np.expand_dims(image_ph_mask, axis=2)
# robust recursive envelope
# up -> down
for i in range(1, image_ph_mask.shape[0]-1):
d = np.abs(image_ph_mask[i-1,:,:] - image_ph_mask[i+1,:,:]) # diff
d = lut_a[(d * lut_a_max).astype(int)]
image_ph_mask[i,:,:] = ((image_ph_mask[i,:,:] * d) +
(image_ph_mask[i-1,:,:] * (1-d)))
# left -> right
for j in range(1, image_ph_mask.shape[1]-1):
d = np.abs(image_ph_mask[:,j-1,:] - image_ph_mask[:,j+1,:]) # diff
d = lut_a[(d * lut_a_max).astype(int)]
image_ph_mask[:,j,:] = ((image_ph_mask[:,j,:] * d) +
(image_ph_mask[:,j-1,:] * (1-d)))
# down -> up
for i in range(image_ph_mask.shape[0]-2, 1, -1):
d = np.abs(image_ph_mask[i-1,:,:] - image_ph_mask[i+1,:,:]) # diff
d = lut_a[(d * lut_a_max).astype(int)]
image_ph_mask[i,:,:] = ((image_ph_mask[i,:,:] * d) +
(image_ph_mask[i+1,:,:] * (1-d)))
# right -> left
for j in range(image_ph_mask.shape[1]-2, 1, -1):
d = np.abs(image_ph_mask[:,j-1,:] - image_ph_mask[:,j+1,:]) # diff
d = lut_b[(d * lut_b_max).astype(int)]
image_ph_mask[:,j,:] = ((image_ph_mask[:,j,:] * d) +
(image_ph_mask[:,j+1,:] * (1-d)))
# up -> down
for i in range(1, image_ph_mask.shape[0]-1):
d = np.abs(image_ph_mask[i-1,:,:] - image_ph_mask[i+1,:,:]) # diff
d = lut_b[(d * lut_b_max).astype(int)]
image_ph_mask[i,:,:] = ((image_ph_mask[i,:,:] * d) +
(image_ph_mask[i-1,:,:] * (1-d)))
# convert back to 2D if grayscale is needed
if grayscale_out is True:
image_ph_mask = np.squeeze(image_ph_mask)
if verbose is True:
plt.figure()
plt.subplot(1,2,1)
plt.imshow(image)
plt.title('Input image')
plt.axis('off')
plt.subplot(1,2,2)
if grayscale_out is True:
plt.imshow(image_ph_mask, cmap='gray', vmin=0, vmax=1)
else:
plt.imshow(image_ph_mask, vmin=0, vmax=1)
plt.title('Photometric mask')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('Estimation of photometric mask')
plt.show()
return image_ph_mask
def blend_expoures(
exposure_list,
threshold_dark=0.35,
threshold_bright=0.65,
verbose=False
):
'''
---------------------------------------------------------------------------
Blend a collection of exposures to a single image
---------------------------------------------------------------------------
Function to blend a list of image exposures, using illumination estimation
across 2 spatial scales.
Based on the following paper:
Vonikakis, V., Bouzos, O. & Andreadis, I. (2011). Multi-Exposure Image
Fusion Based on Illumination Estimation, SIPA2011 (pp.135-142), Greece.
INPUTS
------
exposure_list: list of numpy image arrays
List of numpy arrays (image exposures) which will be blended. Arrays
can be either grayscale, or color (3 channels).
threshold_dark: float in the interval [0,1]
Lower threshold for the membership function which will be applied to
the brightest exposure (long exposure). See above paper for more info.
threshold_dark < threshold_bright
threshold_bright: float in the interval [0,1]
Higher threshold for the membership function which will be applied to
the darkest exposure (short exposure). See above paper for more info.
threshold_bright > threshold_dark
verbose: boolean
Display outputs.
OUTPUT
------
exposure_out: numpy array, float [0,1]
Output image of the blended exposures. If input images are grayscale,
exposure_out is also grayscale. If input images are color, then
exposure_out is also color.
'''
# internal constants
SCALE_COARSE = 0.6 # [0,1], 0->fine, 1->coarse
SCALE_FINE = 0.2 # [0,1], 0->fine, 1->coarse
LUMINANCE_MIDDLE = 0.5 # middle of the luminance scale in [0,1]
GAMA_MAX = 2 # max gama to be used for darkening images
GAMA_MIN = 0.2 # min gama to be used for brightening images
LUT_RESOLUTION = 256
total_exposures = len(exposure_list)
# color or grayscale
if len(exposure_list[0].shape) > 2: # check the 1st image of the list
color_exposures = True
else:
color_exposures = False
#--- sort exposures from darkest to brightest
exposure_list_gray = []
mean_luminance_list = []
if color_exposures is True:
exposure_list_red = []
exposure_list_green = []
exposure_list_blue = []
for image in exposure_list:
image_gray = rgb2gray(image)
exposure_list_gray.append(image_gray) # grayscale
mean_luminance_list.append(image_gray.mean()) # mean luminance
if color_exposures is True:
exposure_list_red.append(img_as_float(image[:,:,0])) # red
exposure_list_green.append(img_as_float(image[:,:,1])) # green
exposure_list_blue.append(img_as_float(image[:,:,2])) # blue
# sort according to mean luminance
indx_lum_ascending = sorted(
range(len(mean_luminance_list)),
key=lambda i: mean_luminance_list[i]
)
if verbose is True:
print('Darkest to brightest exposure sequence:', indx_lum_ascending)
# convert into a numpy array of hight x width x number of exposures
# (the 3rd dimension has the separate grayscale or color exposures)
exposure_array_gray = np.array(exposure_list_gray)
exposure_array_gray = np.moveaxis(exposure_array_gray, 0, -1)
exposure_array_gray = exposure_array_gray[:,:,indx_lum_ascending]
if color_exposures is True:
exposure_array_red = np.array(exposure_list_red)
exposure_array_red = np.moveaxis(exposure_array_red, 0, -1)
exposure_array_red = exposure_array_red[:,:,indx_lum_ascending]
exposure_array_green = np.array(exposure_list_green)
exposure_array_green = np.moveaxis(exposure_array_green, 0, -1)
exposure_array_green = exposure_array_green[:,:,indx_lum_ascending]
exposure_array_blue = np.array(exposure_list_blue)
exposure_array_blue = np.moveaxis(exposure_array_blue, 0, -1)
exposure_array_blue = exposure_array_blue[:,:,indx_lum_ascending]
#--- generate illumination estimation in 2 spatial scales
illumination_coarse = get_photometric_mask(
exposure_array_gray.copy(),
smoothing=SCALE_COARSE,
grayscale_out=False, # estimaste each channel separately
verbose=False)
illumination_fine = get_photometric_mask(
exposure_array_gray.copy(),
smoothing=SCALE_FINE,
grayscale_out=False, # estimaste each channel separately
verbose=False)
# min max normalization for each exposure.
# make sure that each exposure has a 0 and 1 somewhere
for i in range(total_exposures):
illumination_coarse[:,:,i] = rescale_intensity(
illumination_coarse[:,:,i],
in_range='image',
out_range='dtype'
)
illumination_fine[:,:,i] = rescale_intensity(
illumination_fine[:,:,i],
in_range='image',
out_range='dtype'
)
#--- Autoadjusting extreme exposures
# (This would be better if done in a data-driven way)
# if darkest exposure is too bright, darken it
# if brightest exposure is too dark, brighten it
# darkest: if mean_lum>0.5 (too bright)
# scale gamma linearly in the interval [1, GAMA_MAX]
mean_lum = illumination_coarse[:,:,0].mean()
if mean_lum > LUMINANCE_MIDDLE:
gamma_new = map_value(
mean_lum,
range_in=(LUMINANCE_MIDDLE,1),
range_out=(1,GAMA_MAX)
)
if verbose:
print(
'Darkest coarse exposure too bright! Applying gamma:',
gamma_new
)
illumination_coarse[:,:,0] = adjust_gamma(
image = illumination_coarse[:,:,0],
gamma = gamma_new
)
mean_lum = illumination_fine[:,:,0].mean()
if mean_lum > LUMINANCE_MIDDLE:
gamma_new = map_value(
mean_lum,
range_in=(LUMINANCE_MIDDLE,1),
range_out=(1,GAMA_MAX)
)
if verbose:
print(
'Darkest fine exposure too bright! Applying gamma:',
gamma_new
)
illumination_fine[:,:,0] = adjust_gamma(
image = illumination_fine[:,:,0],
gamma = gamma_new
)
# brightest: if mean_lum<0.5 (too dark)
# scale gamma linearly in the interval [GAMA_MIN, 1]
mean_lum = illumination_coarse[:,:,-1].mean()
if mean_lum < LUMINANCE_MIDDLE:
gamma_new = map_value(
mean_lum,
range_in=(0,LUMINANCE_MIDDLE),
range_out=(GAMA_MIN,1)
)
if verbose:
print(
'Brightest coarse exposure too dark! Applying gamma:',
gamma_new
)
illumination_coarse[:,:,-1] = adjust_gamma(
image = illumination_coarse[:,:,-1],
gamma = gamma_new
)
mean_lum = illumination_fine[:,:,-1].mean()
if mean_lum < LUMINANCE_MIDDLE:
gamma_new = map_value(
mean_lum,
range_in=(0,LUMINANCE_MIDDLE),
range_out=(GAMA_MIN,1)
)
if verbose:
print(
'Brightest fine exposure too dark! Applying gamma:',
gamma_new
)
illumination_fine[:,:,-1] = adjust_gamma(
image = illumination_fine[:,:,-1],
gamma = gamma_new
)
#--- Apply membership functions to illumination to get exposure weights
# generate membership function LUTs
weights_lower, weights_mid, weights_upper = get_membership_luts(
resolution=LUT_RESOLUTION,
lower_threshold=threshold_dark, # defines lower cutofd
upper_threshold=threshold_bright, # defines upper cutofd
verbose=verbose
)
lut_resolution = len(weights_lower) - 1
weights_coarse = np.zeros(illumination_coarse.shape, dtype=float)
weights_coarse[:,:,0] = (weights_lower[(illumination_coarse[:,:,0] *
lut_resolution).astype(int)])
weights_coarse[:,:,1:-1] = (weights_mid[(illumination_coarse[:,:,1:-1] *
lut_resolution).astype(int)])
weights_coarse[:,:,-1] = (weights_upper[(illumination_coarse[:,:,-1] *
lut_resolution).astype(int)])
weights_fine = np.zeros(illumination_fine.shape, dtype=float)
weights_fine[:,:,0] = (weights_lower[(illumination_fine[:,:,0] *
lut_resolution).astype(int)])
weights_fine[:,:,1:-1] = (weights_mid[(illumination_fine[:,:,1:-1] *
lut_resolution).astype(int)])
weights_fine[:,:,-1] = (weights_upper[(illumination_fine[:,:,-1] *
lut_resolution).astype(int)])
#TODO: apply local contrast enhancement to the exposure images, 2 times
# (one for each illumination scale)
#--- Weighted average of exposures based on the exposure weights
# grayscale
exposure_coarse = weights_coarse * exposure_array_gray
exposure_coarse = (np.sum(exposure_coarse, axis=2) /
np.sum(weights_coarse, axis=2))
exposure_fine = weights_fine * exposure_array_gray
exposure_fine = (np.sum(exposure_fine, axis=2) /
np.sum(weights_fine, axis=2))
exposure_out_gray = (exposure_coarse + exposure_fine) / 2
exposure_out = exposure_out_gray
if color_exposures is True:
# red
exposure_coarse_red = weights_coarse * exposure_array_red
exposure_coarse_red = (np.sum(exposure_coarse_red, axis=2) /
np.sum(weights_coarse, axis=2))
exposure_fine_red = weights_fine * exposure_array_red
exposure_fine_red = (np.sum(exposure_fine_red, axis=2) /
np.sum(weights_fine, axis=2))
exposure_out_red = (exposure_coarse_red + exposure_fine_red) / 2
# green
exposure_coarse_green = weights_coarse * exposure_array_green
exposure_coarse_green = (np.sum(exposure_coarse_green, axis=2) /
np.sum(weights_coarse, axis=2))
exposure_fine_green = weights_fine * exposure_array_green
exposure_fine_green = (np.sum(exposure_fine_green, axis=2) /
np.sum(weights_fine, axis=2))
exposure_out_green = (exposure_coarse_green + exposure_fine_green) / 2
# blue
exposure_coarse_blue = weights_coarse * exposure_array_blue
exposure_coarse_blue = (np.sum(exposure_coarse_blue, axis=2) /
np.sum(weights_coarse, axis=2))
exposure_fine_blue = weights_fine * exposure_array_blue
exposure_fine_blue = (np.sum(exposure_fine_blue, axis=2) /
np.sum(weights_fine, axis=2))
exposure_out_blue = (exposure_coarse_blue + exposure_fine_blue) / 2
# combine all blended color channels to one image
exposure_out_color = np.zeros(
(exposure_out_gray.shape[0], exposure_out_gray.shape[1], 3),
dtype=float
)
exposure_out_color[:,:,0] = exposure_out_red
exposure_out_color[:,:,1] = exposure_out_green
exposure_out_color[:,:,2] = exposure_out_blue
exposure_out = exposure_out_color
#--- Visualizations
if verbose is True:
# display intermediate stages of the method
plt.figure()
for i in range(total_exposures):
plt.subplot(6,total_exposures,i+1)
plt.imshow(exposure_array_gray[:,:,i], cmap='gray')
plt.title('Exposure ' + str(i))
plt.axis('off')
plt.subplot(6,total_exposures,i+1+total_exposures)
plt.imshow(illumination_coarse[:,:,i], cmap='gray')
plt.title('ill.coarse ' + str(i))
plt.axis('off')
plt.subplot(6,total_exposures,i+1+(total_exposures*2))
plt.imshow(illumination_fine[:,:,i], cmap='gray')
plt.title('ill.fine ' + str(i))
plt.axis('off')
plt.subplot(6,total_exposures,i+1+(total_exposures*3))
plt.imshow(weights_coarse[:,:,i], cmap='gray')
plt.title('W.coarse ' + str(i))
plt.axis('off')
plt.subplot(6,total_exposures,i+1+(total_exposures*4))
plt.imshow(weights_fine[:,:,i], cmap='gray')
plt.title('W.fine ' + str(i))
plt.axis('off')
plt.subplot(6,total_exposures,1+(total_exposures*5))
plt.imshow(exposure_coarse, cmap='gray')
plt.title('Coarse blended')
plt.axis('off')
plt.subplot(6,total_exposures,2+(total_exposures*5))
plt.imshow(exposure_fine, cmap='gray')
plt.title('Fine blended')
plt.axis('off')
plt.subplot(6,total_exposures,3+(total_exposures*5))
plt.imshow(exposure_out_gray, cmap='gray')
plt.title('Final blend')
plt.axis('off')
plt.suptitle('List of exposures')
plt.tight_layout()
plt.tight_layout()
plt.show()
# display final color result
plt.figure()
grid = plt.GridSpec(total_exposures, total_exposures)
if color_exposures is False:
cmap = 'gray'
else:
cmap = None
for i in range(total_exposures):
plt.subplot(grid[0,i])
plt.imshow(exposure_list[indx_lum_ascending[i]], cmap=cmap)
plt.title('Exposure ' + str(i))
plt.axis('off')
plt.subplot(grid[1:,:])
plt.imshow(exposure_out, cmap=cmap)
plt.title('Final blend')
plt.axis('off')
plt.tight_layout()
plt.suptitle('Full color blend')
plt.show()
return exposure_out
def apply_local_contrast_enhancement(
image,
image_ph_mask,
degree=1.5,
verbose=False):
'''
---------------------------------------------------------------------------
Adjust local contrast in an image
---------------------------------------------------------------------------
Increase or decrease the level of local details (local contrast) in an
image. Details are defined as deviations from the local neighborhood
provided by the photometric mask. Dark regions receive also a boost in
local contrast.
INPUTS
------
image: numpy array of WxH of float [0,1]
Input grayscale image.
image_ph_mask: numpy array of WxH of float [0,1]
Grayscale image whose values represent the neighborhood of the pixels
of the input image. Usually, this image some type of edge aware
filtering, such as bilateral filtering, robust recursive envelopes etc.
degree: float [0,inf].
How to change the local contrast.
0: total attenuation of details.
<1: attenuation of details
1: details unchanged
>1: increased local details
verbose: boolean
Display outputs.
OUTPUT
------
image_out: numpy array of WxH of float [0,1]
Output image with adjusted local contrast.
'''
DARK_BOOST = 0.2
THRESHOLD_DARK_TONES = 100 / 255
detail_amplification_global = degree
image_details = image - image_ph_mask # image details
# special treatment for dark regions
detail_amplification_local = image_ph_mask / THRESHOLD_DARK_TONES
detail_amplification_local[detail_amplification_local>1] = 1
detail_amplification_local = ((1 - detail_amplification_local) *
DARK_BOOST) + 1 # [1, 1.2]
# apply all detail adjustements
image_details = (image_details *
detail_amplification_global *
detail_amplification_local)
# add details back to the local neighborhood
image_out = image_ph_mask + image_details
# stay within range
image_out = np.clip(a=image_out, a_min=0, a_max=1, out=image_out)
if verbose is True:
plt.figure()
plt.subplot(1,3,1)
plt.imshow(image, cmap='gray', vmin=0, vmax=1)
plt.title('Input image')
plt.axis('off')
plt.subplot(1,3,2)
plt.imshow(image_ph_mask, cmap='gray', vmin=0, vmax=1)
plt.title('Ph. mask')
plt.axis('off')
plt.subplot(1,3,3)
plt.imshow(image_out, cmap='gray', vmin=0, vmax=1)
plt.title('Output')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('Local contrast enhancement [x' + str(degree) + ']')
plt.show()
return image_out
def apply_spatial_tonemapping(
image,
image_ph_mask,
mid_tone=0.5,
tonal_width=0.5,
areas_dark=0.5,
areas_bright=0.5,
preserve_tones = True,
verbose=True):
'''
---------------------------------------------------------------------------
Apply spatially variable tone mapping based on the local neighborhood
---------------------------------------------------------------------------
Applies different tone mapping curves in each pixel based on its surround.
For surround, the photometric mask is used. Alternatively, other filters
could be used, like gaussian, bilateral filter, edge-avoiding wavelets etc.
Dark pixels are brightened, bright pixels are darkened, and pixels in the
mid_tonedle of the tone range are minimally affected. More information
about the technique can be found in the following papers:
Related publications:
Vonikakis, V., Andreadis, I., & Gasteratos, A. (2008). Fast centre-surround
contrast modification. IET Image processing 2(1), 19-34.
Vonikakis, V., Winkler, S. (2016). A center-surround framework for spatial
image processing. Proc. IS&T Human Vision & Electronic Imaging.
INPUTS
------
image: numpy array of WxH of float [0,1]
Input grayscale image with values in the interval [0,1].
image_ph_mask: numpy array of WxH of float [0,1]
Grayscale image whose values represent the neighborhood of the pixels
of the input image. Usually, this image some type of edge aware
filtering, such as bilateral filtering, robust recursive envelopes etc.
mid_tone: float [0,1]
The mid point between the 'dark' and 'bright' tones. This is equivalent
to a pixel value [0,255], but in the interval [0,1].
tonal_width: float [0,1]
The range of pixel values that will be affected by the correction.
Lower values will localize the enhancement only in a narrow range of
pixel values, whereas for higher values the enhancement will extend to
a greater range of pixel values.
areas_dark: float [0,1]
Degree of enhencement in the dark image areas (0 = no enhencement)
areas_bright: float [0,1]
Degree of enhencement in the bright image areas (0 = no enhencement)
preserve_tones: boolean
Whether or not to preserve well-exposed tones around the middle of the
range.
verbose: boolean
Display outputs.
OUTPUT
------
image_tonemapped: numpy array of WxH of float [0,1]
Tonemapped grayscale image.
'''
# defining parameters
EPSILON = 1 / 256
# adjust range and non-linear response of parameters
mid_tone = map_value(
value=mid_tone,
range_in=(0,1),
range_out=(0,1),
invert=False,
non_lin_convex=None,
non_lin_concave=None
)
tonal_width = map_value(
value=tonal_width,
range_in=(0,1),
range_out=(EPSILON,1),
invert=False,
non_lin_convex=0.1,
non_lin_concave=None
)
areas_dark = map_value(
value=areas_dark,
range_in=(0,1),
range_out=(0,5),
invert=True,
non_lin_convex=0.05,
non_lin_concave=None
)
areas_bright = map_value(
value=areas_bright,
range_in=(0,1),
range_out=(0,5),
invert=True,
non_lin_convex=0.05,
non_lin_concave=None
)
# spatial tone-mapping
# lower tones (below mid_tone level)
image_lower = image.copy()
image_lower[image_lower>=mid_tone] = 0
alpha = (image_ph_mask ** 2) / tonal_width
tone_continuation_factor = mid_tone / (mid_tone + EPSILON - image_ph_mask)
alpha = alpha * tone_continuation_factor + areas_dark
image_lower = (image_lower * (alpha + 1)) / (alpha + image_lower)
# upper tones (above mid_tone level)
image_upper = image.copy()
image_upper[image_upper<mid_tone] = 0
image_ph_mask_inv = 1 - image_ph_mask
alpha = (image_ph_mask_inv ** 2) / tonal_width
tone_continuation_factor = mid_tone / ((1 - mid_tone) - image_ph_mask_inv)
alpha = alpha * tone_continuation_factor + areas_bright
image_upper = (image_upper * alpha) / (alpha + 1 - image_upper)
image_tonemapped = image_lower + image_upper
if preserve_tones is True:
preservation_degree = np.abs(0.5 - image_ph_mask) / 0.5 # 0: near 0.5
# preservation_degree = ((1 + 0.3) * preservation_degree) / (0.3 + preservation_degree)
image_tonemapped = (preservation_degree * image_tonemapped +
(1-preservation_degree) * image)
if verbose is True:
plt.figure()
plt.subplot(2,2,1)
plt.imshow(image, cmap='gray', vmin=0, vmax=1)
plt.title('Input image')
plt.axis('off')
plt.subplot(2,2,3)
plt.imshow(image_lower, cmap='gray', vmin=0, vmax=1)
plt.title('Image lower')
plt.axis('off')
plt.subplot(2,2,4)
plt.imshow(image_upper, cmap='gray', vmin=0, vmax=1)
plt.title('Image upper')
plt.axis('off')
plt.subplot(2,2,2)
plt.imshow(image_tonemapped, cmap='gray', vmin=0, vmax=1)
plt.title('Image tonemapped')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('Spatial tone mapping')
plt.show()
return image_tonemapped
def srgb_to_linear(image_srgb, verbose=False):
'''
---------------------------------------------------------------------------
Transform an image from sRGB color space to linear
---------------------------------------------------------------------------
The function undos the main non-linearities associated with the sRGB color
space, in order to approximate a linear color response. Note that the
linear image output will look darker, because the gamma correction will be
undone. The transformation formulas can be found in the EasyRGB website:
https://www.easyrgb.com/en/math.php
Note that the formulas may look slightly different. This is because they
have been altered in order to implement them in a vectorized way, avoiding
for loops. As such, an image is partitioned in 2 parts image_upper and
image_lower, which implement separate parts of the piece-wise color
transformation formula.
INPUTS
------
image_srgb: numpy array of WxHx3 of uint8 [0,255]
Input color image with values in the interval [0,255]. Assuming that
it is encoded on the sRGB color space. The code will still work if the
input image is grayscale or within [0,1] range.
verbose: boolean
Display outputs.
OUTPUT
------
image_linear: numpy array of WxHx3 of float [0,1]
Output color linear image with values in the interval [0,1]. Gamma has
been removed, so it looks darker.
'''
# dealing with different input dimensions
dimensions = len(image_srgb.shape)
if dimensions == 1:
image_srgb = np.expand_dims(image_srgb, axis=2) # make a 3rd dimension
image_srgb = img_as_float(image_srgb) # [0,1]
# lower part of the piecewise formula
image_lower = image_srgb.copy()
image_lower[image_lower > 0.04045] = 0
image_lower = image_lower / 12.92
# upper part of the piecewise formula
image_upper = image_srgb.copy()
image_upper = image_upper + 0.055
image_upper[image_upper <= (0.04045+0.055)] = 0
image_upper = image_upper / 1.055
image_upper = image_upper ** 2.4
image_linear = image_lower + image_upper # combine into the final result
if verbose is True:
plt.figure()
plt.subplot(1,2,1)
plt.imshow(image_srgb, vmin=0, vmax=1)
plt.title('Image sRGB')
plt.axis('off')
plt.subplot(1,2,2)
plt.imshow(image_linear, vmin=0, vmax=1)
plt.title('Image linear')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('sRGB -> linear space')
plt.show()
return image_linear
def linear_to_srgb(image_linear, verbose=False):
'''
---------------------------------------------------------------------------
Transform an image from linear to sRGB color space
---------------------------------------------------------------------------
The function re-applies the main non-linearities associated with the sRGB
color space. The transformation formula can be found in EasyRGB website:
https://www.easyrgb.com/en/math.php
Note that the formulas may look slightly different. This is because they
have been altered in order to implement them in a vectorized way, avoiding
for loops. As such, an image is partitioned in 2 parts image_upper and
image_lower, which implement separate parts of the piece-wise color
transformation formula.
INPUTS
------
image_linear: numpy array of WxHx3 of float [0,1]
Input color image with values in the interval [0,1].
verbose: boolean
Display outputs.
OUTPUT
------
image_srgb: numpy array of WxHx3 of uint8 [0,255]
Output color sRGB image with values in the interval [0,255].
'''
# dealing with different input dimensions
dimensions = len(image_linear.shape)
if dimensions == 1:
image_linear = np.expand_dims(image_linear, axis=2) # 3rd dimension
image_linear = img_as_float(image_linear) # [0,1]
# lower part of the piecewise formula
image_lower = image_linear.copy()
image_lower[image_lower > 0.0031308] = 0
image_lower = image_lower * 12.92
# upper part of the piecewise formula
image_upper = image_linear.copy()
image_upper[image_upper <= 0.0031308] = 0
image_upper = image_upper ** (1/2.4)
image_upper = image_upper * 1.055
image_upper = image_upper - 0.055
image_srgb = image_lower + image_upper
image_srgb = np.clip(a=image_srgb, a_min=0, a_max=1, out=image_srgb)
if verbose is True:
plt.figure()
plt.subplot(1,2,1)
plt.imshow(image_linear, vmin=0, vmax=1)
plt.title('Image linear')
plt.axis('off')
plt.subplot(1,2,2)
plt.imshow(image_srgb, vmin=0, vmax=1)
plt.title('Image sRGB')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('Linear space -> sRGB')
plt.show()
return (image_srgb * 255).astype(np.uint8)
def transfer_graytone_to_color(image_color, image_graytone, verbose=False):
'''
---------------------------------------------------------------------------
Transfer grayscale tones to a color image
---------------------------------------------------------------------------
Transfers the tones of a guide grayscale image to the color version of the
same image, by using linear color ratios. It first brings the image from
the sRGB color space back to the linear color space. It estimates color
ratios of the grayscale color image with the tone-mapped grayscale guide
image. It then applies the color ratios on the 3 color channels. Finally,
it brings back the image to the sRGB color space (gamma corrected). Is the
input image is in another color space (Adobe RGB), a different
transformation could be used. However, results will not be that much
different.
Related publication:
Chengho Hsin, Zong Wei Lee, Zheng Zhan Lee, and Shaw-Jyh Shin, "Color
preservation for tone reproduction and image enhancement", Proc. SPIE 9015,
Color Imaging XIX, 2014
INPUTS
------
image_color: numpy array of WxHx3 of uint8 [0,255]
Input color image.
image_graytone: numpy array of WxH of float [0,1]
Grayscale version of the image_color which has been tonemapped and it
will be used as a guide to transfer the same tonemapping to the color
image.
verbose: boolean
Display outputs.
OUTPUT
------
image_colortone: numpy array of WxHx3 of uint8 [0,255]
Output color image with transfered tonemapping.
'''
EPSILON = 1 / 256
# bring both color and graytone to linear space
image_color_linear = srgb_to_linear(image_color.copy(), verbose=False)
image_graytone_linear = srgb_to_linear(image_graytone.copy(),verbose=False)
image_gray_linear = rgb2gray(image_color_linear.copy())
image_gray_linear[image_gray_linear==0] = EPSILON # for the division later
# tone ratio of linear images: improved/original
tone_ratio = image_graytone_linear / image_gray_linear
# tone_ratio[np.isinf(tone_ratio)] = 0
# tone_ratio[np.isnan(tone_ratio)] = 0
# apply the tone ratios to the color image
image_colortone_linear = image_color_linear * np.dstack([tone_ratio] * 3)
# make sure it's within limits
image_colortone_linear = np.clip(
a=image_colortone_linear,
a_min=0,
a_max=1,
out=image_colortone_linear
)
# bring back to gamma-corrected sRGB space for visualization
image_colortone = linear_to_srgb(image_colortone_linear, verbose=False)
# display results
if verbose is True:
plt.figure()
plt.subplot(2,4,1)
plt.imshow(image_color, vmin=0, vmax=255)
plt.title('Color')
plt.axis('off')
plt.subplot(2,4,5)
plt.imshow(image_color_linear, vmin=0, vmax=1)
plt.title('Color linear')
plt.axis('off')
plt.subplot(2,4,2)
plt.imshow(image_graytone, cmap='gray', vmin=0, vmax=1)
plt.title('Graytone')
plt.axis('off')
plt.subplot(2,4,6)
plt.imshow(image_graytone_linear, cmap='gray', vmin=0, vmax=1)
plt.title('Graytone linear')
plt.axis('off')
plt.subplot(2,4,7)
plt.imshow(tone_ratio, cmap='gray')
plt.title('Tone ratios')
plt.axis('off')
plt.subplot(2,4,4)
plt.imshow(image_colortone, vmin=0, vmax=255)
plt.title('Colortone')
plt.axis('off')
plt.subplot(2,4,8)
plt.imshow(image_colortone_linear, vmin=0, vmax=1)
plt.title('Colortone linear')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('Transfering gray tones to color')
plt.show()
return image_colortone
def change_color_saturation(
image_color,
image_ph_mask=None,
sat_degree=1.5,
verbose=False):
'''
---------------------------------------------------------------------------
Adjust color saturation of an image
---------------------------------------------------------------------------
Increase or decrease the saturation (vibrance) of colors in an image. This
implements a simpler approach rather than using the HSV color space to
adjust S. In my experiments HSV-based saturation adjustment was not as good
and it exhibited some kind of 'color noise'. This approach is aesthetically
better. The use of photometric_mask is optional, in case you would like to
treat dark areas (where saturation is usually lower) differently.
INPUTS
------
image_color: numpy array of WxHx3 of float [0,1]
Input color image.
image_ph_mask: numpy array of WxH of float [0,1] or None
Grayscale image whose values represent the neighborhood of the pixels
of the input image. If None, saturation adjustment is applied globally
to all pixels. If not None, then dark regions are treated differently
and get an additional boost in saturation.
sat_degree': float [0,inf].
How to change the color saturation. 0: no color (grayscale),
<1: reduced color saturation, 1: color saturation unchanged
>1: increased color saturation
verbose: boolean
Display outputs.
OUTPUT
------
image_new_sat: numpy array of WxHx3 of float [0,1]
Output image with adjusted saturation.
'''
LOCAL_BOOST = 0.2
THRESHOLD_DARK_TONES = 100 / 255
#TODO: return the same image type
image_color = img_as_float(image_color) # [0,1]
# define gray scale
image_gray = (image_color[:,:,0] +
image_color[:,:,1] +
image_color[:,:,2]) / 3
image_gray = np.dstack([image_gray] * 3) # grayscale with 3 channels
image_delta = image_color - image_gray # deviations from gray
# defining local color amplification degree
if image_ph_mask is not None:
detail_amplification_local = image_ph_mask / THRESHOLD_DARK_TONES
detail_amplification_local[detail_amplification_local>1] = 1
detail_amplification_local = ((1 - detail_amplification_local) *
LOCAL_BOOST) + 1 # [1, 1.2]
detail_amplification_local = np.dstack(
[detail_amplification_local] * 3) # 3 channels
else:
detail_amplification_local = 1
image_new_sat = (image_gray +
image_delta * sat_degree * detail_amplification_local)
image_new_sat = np.clip(
a=image_new_sat,
a_min=0,
a_max=1,
out=image_new_sat
)
if verbose is True:
plt.figure()
plt.subplot(1,2,1)
plt.imshow(image_color, vmin=0, vmax=1)
plt.title('Input image')
plt.axis('off')
plt.subplot(1,2,2)
plt.imshow(image_new_sat, vmin=0, vmax=1)
plt.title('New saturation [x' + str(sat_degree) + ']')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('Color saturation adjustment')
plt.show()
return image_new_sat
def correct_colors(image, verbose):
'''
---------------------------------------------------------------------------
Correct image colors (remove color casts)
---------------------------------------------------------------------------
Implements a simple color correction using the Gray World Color Assumption
and White Point Correction.
Related publication:
Vonikakis, V., Arapakis, I. & Andreadis, I. (2011). Combining Gray-World
assumption, White-Point correction and power transformation for automatic
white balance. International Workshop on Advanced Image Technology (IWAIT),
paper number 1569353295, Jakarta Indonesia.
INPUTS
------
image: numpy array of WxHx3 of uint8 [0,255]
Input color image.
verbose: boolean
Display outputs.
OUTPUT
------
image_out: numpy array of WxHx3 of float [0,1]
Output image with adjusted colors.
'''
image_out = img_as_float(image.copy()) # [0,1]
# # simple gray world color correction
# image_out[:,:,0] = (image_out[:,:,0] / image_out[:,:,0].mean()) * 0.5
# image_out[:,:,1] = (image_out[:,:,1] / image_out[:,:,1].mean()) * 0.5
# image_out[:,:,2] = (image_out[:,:,2] / image_out[:,:,2].mean()) * 0.5
# mean of all channels
image_mean = (image_out[:,:,0].mean() +
image_out[:,:,1].mean() +
image_out[:,:,2].mean()) / 3
# logarithm base to which each channel will be raised
base_r = image_out[:,:,0].mean() / image_out[:,:,0].max()
base_g = image_out[:,:,1].mean() / image_out[:,:,1].max()
base_b = image_out[:,:,2].mean() / image_out[:,:,2].max()
# the power to which each channel will be raised
power_r = math.log(image_mean, base_r)
power_g = math.log(image_mean, base_g)
power_b = math.log(image_mean, base_b)
# separately applying different color correction powers to each channel
image_out[:,:,0] = (image_out[:,:,0] / image_out[:,:,0].max()) ** power_r
image_out[:,:,1] = (image_out[:,:,1] / image_out[:,:,1].max()) ** power_g
image_out[:,:,2] = (image_out[:,:,2] / image_out[:,:,2].max()) ** power_b
if verbose is True:
plt.figure()
plt.subplot(1,2,1)
plt.imshow(image)
plt.title('Input image')
plt.axis('off')
plt.subplot(1,2,2)
plt.imshow(image_out, vmin=0, vmax=1)
plt.title('Corrected colors')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('Gray world color correction')
plt.show()
return image_out
def adjust_brightness(image, degree=0, verbose=False):
'''
---------------------------------------------------------------------------
Apply global tone mapping on a grayscale image
---------------------------------------------------------------------------
Applies a single tone mapping curve in all the pixels of a grayscale image.
Depending on the parameters, the image can be brighten or darken. The set
of curves used are similar to gamma functions, but are inspired from the
Naka-Rushton function and exhibit symmetry and better local contrast. More
information about the technique can be found in the following papers:
Related publications:
Vonikakis, V., Winkler, S. (2016). A center-surround framework for spatial
image processing. Proc. IS&T Human Vision & Electronic Imaging.
INPUTS
------
image: numpy array of WxH of float [0,1]
Input grayscale image with values in the interval [0,1].
degree: float [-1,1]
The strength of the uniform tone mapping function.
[-1,0): darken image. Closer to -1 means more agressive darkening
0: Unchanged tones
(0,1]: brighten image. Closer to 1 means more agressive brightening
verbose: boolean
Display outputs.
OUTPUT
------
image_tonemapped: numpy array of WxH of float [0,1]
Tonemapped grayscale image.
'''
EPSILON = 1 / 256 # what we consider minimum value
# adjust range and non-linear response of parameters
# unpack information: darken or brighten and the degree
if degree > 0:
brighten = True
else:
brighten = False
degree = abs(degree) # [0,1]
alpha = map_value(
value=degree,
range_in=(0,1),
range_out=(0,5), # from the paper: 5x brings close to linear
invert=True, # from the paper
non_lin_convex=0.05, # adding linearity to the response
non_lin_concave=None
)
alpha = alpha + EPSILON # to avoid division by zero
# applying global tone-mapping
if degree != 0:
image_brightness = image.copy()
if brighten is True:
image_brightness = ((image_brightness * (alpha + 1)) /
(alpha + image_brightness))
else:
image_brightness = ((image_brightness * alpha) /
(alpha + 1 - image_brightness))
else: image_brightness = image
if verbose is True:
plt.figure()
plt.subplot(1,2,1)
plt.imshow(image, cmap='gray', vmin=0, vmax=1)
plt.title('Input image')
plt.axis('off')
plt.subplot(1,2,2)
plt.imshow(image_brightness, cmap='gray', vmin=0, vmax=1)
plt.title('Adjusted brightness image')
plt.axis('off')
plt.tight_layout(True)
plt.suptitle('Adjusting brightness')
plt.show()
return image_brightness
def enhance_image(image, parameters, verbose=False):
'''
---------------------------------------------------------------------------
Image enhancement
---------------------------------------------------------------------------
Image enhancement pipeline, with spatial tone mapping, local contrast
enhancement and color saturation adjustment. The 3 steps are fully
decoupled and the user can independently define the enhancement degree of
each stage.
Related publications:
Vonikakis, V., Andreadis, I., & Gasteratos, A. (2008). Fast centre-surround
contrast modification. IET Image processing 2(1), 19-34.
Vonikakis, V., Winkler, S. (2016). A center-surround framework for spatial
image processing. Proc. IS&T Human Vision & Electronic Imaging.
INPUTS
------
image: numpy array of WxHx3 of uint8 [0,255]
Input color image with values in the interval [0,255].
parameters: dictionary
'local_contrast': float [0,inf].
0: total attenuation of details.
<1: attenuation of details
1: details unchanged
>1: increased local details
'mid_tones': float [0,1]
'tonal_width': float [0,1]
'areas_dark': float [0,1]
0: no enhancement
1: strongest enhancement
'areas_bright': float [0,1]
0: no enhancement
1: strongest enhancement
'brightness': float [-1,1]
>=-1: darken image
0: unchanged
<=1: brighten image
'preserve_tones': boolean
'color_correction': boolean
'saturation_degree': float [0,inf].
0: no color (grayscale).
<1: reduced color saturation
1: color saturation unchanged
>1: increased color saturation
verbose: boolean
Display outputs.
OUTPUT
------
image_colortone_saturation: numpy array of WxHx3 of uint8 [0,255]
Output enhanced image.
'''
#TODO: add an automatic parameter estimation stage (machine learning)
# sanity check for type, range and defaults
if 'local_contrast' in parameters:
parameters['local_contrast'] = float(parameters['local_contrast'])
if parameters['local_contrast'] < 0: parameters['local_contrast'] = 0
else: parameters['local_contrast'] = 1.2 # default: slight increase
if 'mid_tones' in parameters:
parameters['mid_tones'] = float(parameters['mid_tones'])
if parameters['mid_tones'] > 1: parameters['mid_tones'] = 1
if parameters['mid_tones'] < 0: parameters['mid_tones'] = 0
else: parameters['mid_tones'] = 0.5 # default: middle of the range
if 'tonal_width' in parameters:
parameters['tonal_width'] = float(parameters['tonal_width'])
if parameters['tonal_width'] > 1: parameters['tonal_width'] = 1
if parameters['tonal_width'] < 0: parameters['tonal_width'] = 0
else: parameters['tonal_width'] = 0.5 # default: middle of the range
if 'areas_dark' in parameters:
parameters['areas_dark'] = float(parameters['areas_dark'])
if parameters['areas_dark'] > 1: parameters['areas_dark'] = 1
if parameters['areas_dark'] < 0: parameters['areas_dark'] = 0
else: parameters['areas_dark'] = 0.2 # default: gentle increase
if 'areas_bright' in parameters:
parameters['areas_bright'] = float(parameters['areas_bright'])
if parameters['areas_bright'] > 1: parameters['areas_bright'] = 1
if parameters['areas_bright'] < 0: parameters['areas_bright'] = 0
else: parameters['areas_bright'] = 0.2 # default: gentle increase
if 'brightness' in parameters:
parameters['brightness'] = float(parameters['brightness'])
if parameters['brightness'] > 1: parameters['brightness'] = 1
if parameters['brightness'] < -1: parameters['brightness'] = -1
else: parameters['brightness'] = 0.1 # default: gentle increase
if 'preserve_tones' in parameters:
parameters['preserve_tones'] = bool(parameters['preserve_tones'])
else: parameters['preserve_tones'] = True # default: preserve tones
if 'color_correction' in parameters:
parameters['color_correction'] = bool(parameters['color_correction'])
else: parameters['color_correction'] = False # default: no correction
if 'saturation_degree' in parameters:
parameters['saturation_degree'] = float(parameters['saturation_degree'])
if parameters['saturation_degree'] < 0: parameters['saturation_degree'] = 0
else: parameters['saturation_degree'] = 1.2 # default: slight increase
# get photometric mask, as a guide for spatial-tone mapping
image_ph_mask = get_photometric_mask(
image=image,
verbose=verbose
)
# increase the local contrast of the grayscale image
image_contrast = apply_local_contrast_enhancement(
image=rgb2gray(image.copy()),
image_ph_mask=image_ph_mask,
degree=parameters['local_contrast'],
verbose=verbose
)
# apply spatial tonemapping on the previous stage
image_tonemapped = apply_spatial_tonemapping(
image=image_contrast,
image_ph_mask=image_ph_mask,
mid_tone=parameters['mid_tones'],
tonal_width=parameters['tonal_width'],
areas_dark=parameters['areas_dark'],
areas_bright=parameters['areas_bright'],
preserve_tones=parameters['preserve_tones'],
verbose=verbose
)
image_brightness = adjust_brightness(
image_tonemapped,
degree=parameters['brightness'],
verbose=verbose
)
# transfer the enhancement on the color image (in the linear color space)
image_colortone = transfer_graytone_to_color(
image_color=image,
image_graytone=image_brightness,
verbose=verbose
)
# apply color correction (if needed)
if parameters['color_correction'] is True:
image_colortone = correct_colors(
image=image_colortone,
verbose=verbose
)
# adjust the color saturation
image_colortone_saturation = change_color_saturation(
image_color=image_colortone,
image_ph_mask=image_ph_mask,
sat_degree=parameters['saturation_degree'],
verbose = verbose,
)
# TODO: add a denoising stage
# display results
if verbose is True:
plt.figure()
plt.subplot(2,3,1)
plt.imshow(image, vmin=0, vmax=255)
plt.title('Input image')
plt.axis('off')
plt.tight_layout()
plt.subplot(2,3,4)
plt.imshow(image_ph_mask, cmap='gray', vmin=0, vmax=1)
plt.title('Photometric mask')
plt.axis('off')
plt.tight_layout()
plt.subplot(2,3,5)
plt.imshow(image_contrast, cmap='gray', vmin=0, vmax=1)
plt.title('Local contrast enhancement')
plt.axis('off')
plt.tight_layout()
plt.subplot(2,3,2)
plt.imshow(image_colortone, vmin=0, vmax=255)
plt.title('Spatial tone mapping')
plt.axis('off')
plt.tight_layout()
plt.subplot(2,3,3)
plt.imshow(image_colortone_saturation, vmin=0, vmax=255)
plt.title('Increased saturation')
plt.axis('off')
plt.tight_layout()
return image_colortone_saturation
# def fuse_exposures(ls_images):
if __name__=="__main__":
filename = "../images/lisbon.jpg"
image = imageio.imread(filename) # load image
# setting up parameters
parameters = {}
parameters['local_contrast'] = 1.5 # 1.5x increase in details
parameters['mid_tones'] = 0.5
parameters['tonal_width'] = 0.5
parameters['areas_dark'] = 0.7 # 70% improvement in dark areas
parameters['areas_bright'] = 0.5 # 50% improvement in bright areas
parameters['saturation_degree'] = 1.2 # 1.2x increase in color saturation
parameters['brightness'] = 0.1 # slight increase in brightness
parameters['preserve_tones'] = True
parameters['color_correction'] = False
image_enhanced = enhance_image(image, parameters, verbose=False)
# display results
plt.figure()
plt.subplot(1,2,1)
plt.imshow(image, vmin=0, vmax=255)
plt.title('Input image')
plt.axis('off')
plt.tight_layout()
plt.subplot(1,2,2)
plt.imshow(image_enhanced, vmin=0, vmax=255)
plt.title('Enhanced image')
plt.axis('off')
plt.tight_layout()
plt.show()
|