File size: 109,087 Bytes
2523bef | 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 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 | """
Copyright (C) 2023 Artifex Software, Inc.
This file is part of PyMuPDF.
PyMuPDF is free software: you can redistribute it and/or modify it under the
terms of the GNU Affero General Public License as published by the Free
Software Foundation, either version 3 of the License, or (at your option)
any later version.
PyMuPDF is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more
details.
You should have received a copy of the GNU Affero General Public License
along with MuPDF. If not, see <https://www.gnu.org/licenses/agpl-3.0.en.html>
Alternative licensing terms are available from the licensor.
For commercial licensing, see <https://www.artifex.com/> or contact
Artifex Software, Inc., 39 Mesa Street, Suite 108A, San Francisco,
CA 94129, USA, for further information.
---------------------------------------------------------------------
Portions of this code have been ported from pdfplumber, see
https://pypi.org/project/pdfplumber/.
The ported code is under the following MIT license:
---------------------------------------------------------------------
The MIT License (MIT)
Copyright (c) 2015, Jeremy Singer-Vine
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
---------------------------------------------------------------------
Also see here: https://github.com/jsvine/pdfplumber/blob/stable/LICENSE.txt
---------------------------------------------------------------------
The porting mainly pertains to files "table.py" and relevant parts of
"utils/text.py" within pdfplumber's repository on Github.
With respect to "text.py", we have removed functions or features that are not
used by table processing. Examples are:
* the text search function
* simple text extraction
* text extraction by lines
Original pdfplumber code does neither detect, nor identify table headers.
This PyMuPDF port adds respective code to the 'Table' class as method '_get_header'.
This is implemented as new class TableHeader with the properties:
* bbox: A tuple for the header's bbox
* cells: A tuple for each bbox of a column header
* names: A list of strings with column header text
* external: A bool indicating whether the header is outside the table cells.
"""
import inspect
import itertools
import string
import html
from collections.abc import Sequence
from contextvars import ContextVar
from dataclasses import dataclass
from operator import itemgetter
import weakref
import pymupdf
from pymupdf import mupdf
# The opt-in refine, cell-span and layout-union stages live in sibling modules,
# re-imported here to keep the public pymupdf.table.* surface unchanged.
# _table_union imports find_tables back from this module, so it is imported
# lazily inside find_tables() instead, to avoid an import cycle.
from pymupdf._table_refine import (
refine_grid_structure,
refine_grid_rows,
_refine_cells_to_grid,
_refine_grid_to_cells,
_refine_page_words,
)
from pymupdf._table_spans import (
resolve_spans,
SpanCell,
_span_select_words_in_rect,
_span_word_line_tuple,
_span_words_to_line_text,
)
# Header semantics and HTML serialization live in the _table_headers sibling.
from pymupdf._table_headers import (
find_header_region,
collapse_cell_ws,
render_table_html,
)
# refine_grid is unused in this module; re-exported for the public
# pymupdf.table.* surface.
from pymupdf._table_refine import refine_grid # noqa: F401 # pylint: disable=unused-import
# -------------------------------------------------------------------
# Start of PyMuPDF interface code
# -------------------------------------------------------------------
# pylint: disable=no-name-in-module
_EDGES_VAR = ContextVar("pymupdf_table_edges", default=None)
_CHARS_VAR = ContextVar("pymupdf_table_chars", default=None)
class _TableStateList:
"""List-like proxy for per-call table extraction state."""
def __init__(self, var):
self._var = var
def _list(self):
value = self._var.get()
if value is None:
value = []
self._var.set(value)
return value
def append(self, item):
return self._list().append(item)
def clear(self):
return self._list().clear()
def extend(self, items):
return self._list().extend(items)
def __bool__(self):
return bool(self._list())
def __getitem__(self, item):
return self._list()[item]
def __iter__(self):
return iter(self._list())
def __len__(self):
return len(self._list())
def __setitem__(self, key, value):
self._list()[key] = value
EDGES = _TableStateList(_EDGES_VAR) # vector graphics from PyMuPDF
CHARS = _TableStateList(_CHARS_VAR) # text characters from PyMuPDF
TEXT_BOLD = mupdf.FZ_STEXT_BOLD
TEXT_STRIKEOUT = mupdf.FZ_STEXT_STRIKEOUT
FLAGS = (
0
| pymupdf.TEXTFLAGS_TEXT
| pymupdf.TEXT_COLLECT_STYLES
| pymupdf.TEXT_ACCURATE_BBOXES
| pymupdf.TEXT_MEDIABOX_CLIP
)
# needed by mupdf function fz_find_table_within_bounds().
TABLE_DETECTOR_FLAGS = (
0
| pymupdf.TEXT_ACCURATE_BBOXES
| pymupdf.TEXT_SEGMENT
| pymupdf.TEXT_COLLECT_VECTORS
| pymupdf.TEXT_MEDIABOX_CLIP
)
white_spaces = set(string.whitespace) # for checking white space only cells
def rect_in_rect(inner, outer):
"""Check whether rectangle 'inner' is fully inside rectangle 'outer'."""
return (
1
and inner[0] >= outer[0]
and inner[1] >= outer[1]
and inner[2] <= outer[2]
and inner[3] <= outer[3]
)
def chars_in_rect(CHARS, rect):
"""Check whether any of the chars in CHAR are inside rectangle 'rect'."""
return any(
1
and rect[0] <= c["x0"]
and c["x1"] <= rect[2]
and rect[1] <= c["y0"]
and rect[3] >= c["y1"]
for c in CHARS
)
def _iou(r1, r2):
"""Compute intersection over union of two rectangles."""
ix = max(0, min(r1[2], r2[2]) - max(r1[0], r2[0]))
iy = max(0, min(r1[3], r2[3]) - max(r1[1], r2[1]))
intersection = ix * iy # intersection area
if not intersection:
return 0
area1 = (r1[2] - r1[0]) * (r1[3] - r1[1])
area2 = (r2[2] - r2[0]) * (r2[3] - r2[1])
return intersection / (area1 + area2 - intersection)
def intersects_words_h(bbox, y, word_rects) -> bool:
"""Check whether any of the words in bbox are cut through by
horizontal line y.
"""
return any(r.y0 < y < r.y1 for r in word_rects if rect_in_rect(r, bbox))
def get_table_dict_from_rect(textpage, rect):
"""Extract MuPDF table structure information from a given rectangle."""
table_dict = {}
pymupdf.extra.make_table_dict(textpage.this.m_internal, table_dict, rect)
return table_dict
def make_table_from_bbox(textpage, word_rects, rect):
"""Detect table structure within a given rectangle."""
cells = [] # table cells as (x0,y0,x1,y1) tuples
# calls fz_find_table_within_bounds
block = get_table_dict_from_rect(textpage, rect)
# No table structure found if not a grid block
if block.get("type") != mupdf.FZ_STEXT_BLOCK_GRID:
return cells
bbox = pymupdf.Rect(block["bbox"]) # resulting table bbox
# lists of (pos,uncertainty) tuples
xpos = sorted(block["xpos"], key=lambda x: x[0])
ypos = sorted(block["ypos"], key=lambda y: y[0])
# maximum uncertainties in x and y directions
xmaxu, ymaxu = block["max_uncertain"]
# Modify ypos to remove uncertain positions, and y positions
# that cut through words.
nypos = []
for y, yunc in ypos:
if yunc > 0: # allow no uncertain y values
continue
if intersects_words_h(bbox, y, word_rects):
continue # allow no y that cuts through words
if nypos and (y - nypos[-1] < 3):
nypos[-1] = y # snap close positions
else:
nypos.append(y)
# New max y uncertainty: 35% of remaining y positions.
# Omit x positions that intersect too many words, otherwise
# only remove x for the affected cells.
ymaxu = max(0, round((len(nypos) - 2) * 0.35))
# Exclude x positions with too high uncertainty
# (we allow more uncertainty in x direction)
nxpos = [x[0] for x in xpos if x[1] <= ymaxu]
if bbox.x1 > nxpos[-1] + 3:
nxpos.append(bbox.x1) # ensure right table border
# Compose cells from the remaining x and y positions.
for i in range(len(nypos) - 1):
row_box = pymupdf.Rect(bbox.x0, nypos[i], bbox.x1, nypos[i + 1])
# Sub-select words in this row and sort them by left coordinate
row_words = sorted(
[r for r in word_rects if rect_in_rect(r, row_box)], key=lambda r: r.x0
)
# Sub-select x values that do not cut through words
this_xpos = [x for x in nxpos if not any(r.x0 < x < r.x1 for r in row_words)]
for j in range(len(this_xpos) - 1):
cell = pymupdf.Rect(this_xpos[j], nypos[i], this_xpos[j + 1], nypos[i + 1])
if not cell.is_empty: # valid cell
cells.append(tuple(cell))
# Add new table to TableFinder tables
return cells
def extract_cells(textpage, cell, markdown=False):
"""Extract text from a rect-like 'cell' as plain or MD styled text.
This function should ultimately be used to extract text from a table cell.
Markdown output will only work correctly if extraction flag bit
TEXT_COLLECT_STYLES is set.
Args:
textpage: A PyMuPDF TextPage object. Must have been created with
TEXTFLAGS_TEXT | TEXT_COLLECT_STYLES.
cell: A tuple (x0, y0, x1, y1) defining the cell's bbox.
markdown: If True, return text formatted for Markdown.
Returns:
A string with the text extracted from the cell.
"""
text = ""
for block in textpage.extractRAWDICT()["blocks"]:
if block["type"] != 0:
continue
block_bbox = block["bbox"]
if (
0
or block_bbox[0] > cell[2]
or block_bbox[2] < cell[0]
or block_bbox[1] > cell[3]
or block_bbox[3] < cell[1]
):
continue # skip block outside cell
for line in block["lines"]:
lbbox = line["bbox"]
if (
0
or lbbox[0] > cell[2]
or lbbox[2] < cell[0]
or lbbox[1] > cell[3]
or lbbox[3] < cell[1]
):
continue # skip line outside cell
if text: # must be a new line in the cell
text += "<br>" if markdown else "\n"
# strikeout detection only works with horizontal text
horizontal = line["dir"] == (0, 1) or line["dir"] == (1, 0)
for span in line["spans"]:
sbbox = span["bbox"]
if (
0
or sbbox[0] > cell[2]
or sbbox[2] < cell[0]
or sbbox[1] > cell[3]
or sbbox[3] < cell[1]
):
continue # skip spans outside cell
# only include chars with more than 50% bbox overlap
span_text = ""
for char in span["chars"]:
this_char = char["c"]
bbox = pymupdf.Rect(char["bbox"])
if abs(bbox & cell) > 0.5 * abs(bbox):
span_text += this_char
elif this_char in white_spaces:
span_text += " "
if not span_text:
continue # skip empty span
if not markdown: # no MD styling
text += span_text
continue
prefix = ""
suffix = ""
if horizontal and span["char_flags"] & TEXT_STRIKEOUT:
prefix += "~~"
suffix = "~~" + suffix
if span["char_flags"] & TEXT_BOLD:
prefix += "**"
suffix = "**" + suffix
if span["flags"] & pymupdf.TEXT_FONT_ITALIC:
prefix += "_"
suffix = "_" + suffix
if span["flags"] & pymupdf.TEXT_FONT_MONOSPACED:
prefix += "`"
suffix = "`" + suffix
if len(span["chars"]) > 2:
span_text = span_text.rstrip()
# if span continues previous styling: extend cell text
if (ls := len(suffix)) and text.endswith(suffix):
text = text[:-ls] + span_text + suffix
else: # append the span with new styling
if not span_text.strip():
text += " "
else:
text += prefix + span_text + suffix
return text.strip()
# -------------------------------------------------------------------
# End of PyMuPDF interface code
# -------------------------------------------------------------------
class UnsetFloat(float):
pass
NON_NEGATIVE_SETTINGS = [
"snap_tolerance",
"snap_x_tolerance",
"snap_y_tolerance",
"join_tolerance",
"join_x_tolerance",
"join_y_tolerance",
"edge_min_length",
"min_words_vertical",
"min_words_horizontal",
"intersection_tolerance",
"intersection_x_tolerance",
"intersection_y_tolerance",
]
TABLE_STRATEGIES = ["lines", "lines_strict", "text", "explicit"]
UNSET = UnsetFloat(0)
DEFAULT_SNAP_TOLERANCE = 3
DEFAULT_JOIN_TOLERANCE = 3
DEFAULT_MIN_WORDS_VERTICAL = 3
DEFAULT_MIN_WORDS_HORIZONTAL = 1
DEFAULT_X_TOLERANCE = 3
DEFAULT_Y_TOLERANCE = 3
DEFAULT_X_DENSITY = 7.25
DEFAULT_Y_DENSITY = 13
bbox_getter = itemgetter("x0", "top", "x1", "bottom")
LIGATURES = {
"ff": "ff",
"ffi": "ffi",
"ffl": "ffl",
"fi": "fi",
"fl": "fl",
"st": "st",
"ſt": "st",
}
def to_list(collection) -> list:
if isinstance(collection, list):
return collection
elif isinstance(collection, Sequence):
return list(collection)
elif hasattr(collection, "to_dict"):
res = collection.to_dict("records") # pragma: nocover
return res
else:
return list(collection)
class TextMap:
"""
A TextMap maps each unicode character in the text to an individual `char`
object (or, in the case of layout-implied whitespace, `None`).
"""
def __init__(self, tuples=None) -> None:
self.tuples = tuples
self.as_string = "".join(map(itemgetter(0), tuples))
def match_to_dict(
self,
m,
main_group: int = 0,
return_groups: bool = True,
return_chars: bool = True,
) -> dict:
subset = self.tuples[m.start(main_group) : m.end(main_group)]
chars = [c for (text, c) in subset if c is not None]
x0, top, x1, bottom = objects_to_bbox(chars)
result = {
"text": m.group(main_group),
"x0": x0,
"top": top,
"x1": x1,
"bottom": bottom,
}
if return_groups:
result["groups"] = m.groups()
if return_chars:
result["chars"] = chars
return result
class WordMap:
"""
A WordMap maps words->chars.
"""
def __init__(self, tuples) -> None:
self.tuples = tuples
def to_textmap(
self,
layout: bool = False,
layout_width=0,
layout_height=0,
layout_width_chars: int = 0,
layout_height_chars: int = 0,
x_density=DEFAULT_X_DENSITY,
y_density=DEFAULT_Y_DENSITY,
x_shift=0,
y_shift=0,
y_tolerance=DEFAULT_Y_TOLERANCE,
use_text_flow: bool = False,
presorted: bool = False,
expand_ligatures: bool = True,
) -> TextMap:
"""
Given a list of (word, chars) tuples (i.e., a WordMap), return a list of
(char-text, char) tuples (i.e., a TextMap) that can be used to mimic the
structural layout of the text on the page(s), using the following approach:
- Sort the words by (doctop, x0) if not already sorted.
- Calculate the initial doctop for the starting page.
- Cluster the words by doctop (taking `y_tolerance` into account), and
iterate through them.
- For each cluster, calculate the distance between that doctop and the
initial doctop, in points, minus `y_shift`. Divide that distance by
`y_density` to calculate the minimum number of newlines that should come
before this cluster. Append that number of newlines *minus* the number of
newlines already appended, with a minimum of one.
- Then for each cluster, iterate through each word in it. Divide each
word's x0, minus `x_shift`, by `x_density` to calculate the minimum
number of characters that should come before this cluster. Append that
number of spaces *minus* the number of characters and spaces already
appended, with a minimum of one. Then append the word's text.
- At the termination of each line, add more spaces if necessary to
mimic `layout_width`.
- Finally, add newlines to the end if necessary to mimic to
`layout_height`.
Note: This approach currently works best for horizontal, left-to-right
text, but will display all words regardless of orientation. There is room
for improvement in better supporting right-to-left text, as well as
vertical text.
"""
_textmap = []
if not len(self.tuples):
return TextMap(_textmap)
expansions = LIGATURES if expand_ligatures else {}
if layout:
if layout_width_chars:
if layout_width:
raise ValueError(
"`layout_width` and `layout_width_chars` cannot both be set."
)
else:
layout_width_chars = int(round(layout_width / x_density))
if layout_height_chars:
if layout_height:
raise ValueError(
"`layout_height` and `layout_height_chars` cannot both be set."
)
else:
layout_height_chars = int(round(layout_height / y_density))
blank_line = [(" ", None)] * layout_width_chars
else:
blank_line = []
num_newlines = 0
words_sorted_doctop = (
self.tuples
if presorted or use_text_flow
else sorted(self.tuples, key=lambda x: float(x[0]["doctop"]))
)
first_word = words_sorted_doctop[0][0]
doctop_start = first_word["doctop"] - first_word["top"]
for i, ws in enumerate(
cluster_objects(
words_sorted_doctop, lambda x: float(x[0]["doctop"]), y_tolerance
)
):
y_dist = (
(ws[0][0]["doctop"] - (doctop_start + y_shift)) / y_density
if layout
else 0
)
num_newlines_prepend = max(
# At least one newline, unless this iis the first line
int(i > 0),
# ... or as many as needed to get the imputed "distance" from the top
round(y_dist) - num_newlines,
)
for i in range(num_newlines_prepend):
if not len(_textmap) or _textmap[-1][0] == "\n":
_textmap += blank_line
_textmap.append(("\n", None))
num_newlines += num_newlines_prepend
line_len = 0
line_words_sorted_x0 = (
ws
if presorted or use_text_flow
else sorted(ws, key=lambda x: float(x[0]["x0"]))
)
for word, chars in line_words_sorted_x0:
x_dist = (word["x0"] - x_shift) / x_density if layout else 0
num_spaces_prepend = max(min(1, line_len), round(x_dist) - line_len)
_textmap += [(" ", None)] * num_spaces_prepend
line_len += num_spaces_prepend
for c in chars:
letters = expansions.get(c["text"], c["text"])
for letter in letters:
_textmap.append((letter, c))
line_len += 1
# Append spaces at end of line
if layout:
_textmap += [(" ", None)] * (layout_width_chars - line_len)
# Append blank lines at end of text
if layout:
num_newlines_append = layout_height_chars - (num_newlines + 1)
for i in range(num_newlines_append):
if i > 0:
_textmap += blank_line
_textmap.append(("\n", None))
# Remove terminal newline
if _textmap[-1] == ("\n", None):
_textmap = _textmap[:-1]
return TextMap(_textmap)
class WordExtractor:
def __init__(
self,
x_tolerance=DEFAULT_X_TOLERANCE,
y_tolerance=DEFAULT_Y_TOLERANCE,
keep_blank_chars: bool = False,
use_text_flow=False,
horizontal_ltr=True, # Should words be read left-to-right?
vertical_ttb=False, # Should vertical words be read top-to-bottom?
extra_attrs=None,
split_at_punctuation=False,
expand_ligatures=True,
):
self.x_tolerance = x_tolerance
self.y_tolerance = y_tolerance
self.keep_blank_chars = keep_blank_chars
self.use_text_flow = use_text_flow
self.horizontal_ltr = horizontal_ltr
self.vertical_ttb = vertical_ttb
self.extra_attrs = [] if extra_attrs is None else extra_attrs
# Note: string.punctuation = '!"#$%&\'()*+,-./:;<=>?@[\\]^_`{|}~'
self.split_at_punctuation = (
string.punctuation
if split_at_punctuation is True
else (split_at_punctuation or "")
)
self.expansions = LIGATURES if expand_ligatures else {}
def merge_chars(self, ordered_chars: list):
x0, top, x1, bottom = objects_to_bbox(ordered_chars)
doctop_adj = ordered_chars[0]["doctop"] - ordered_chars[0]["top"]
upright = ordered_chars[0]["upright"]
direction = 1 if (self.horizontal_ltr if upright else self.vertical_ttb) else -1
matrix = ordered_chars[0]["matrix"]
rotation = 0
if not upright and matrix[1] < 0:
ordered_chars = reversed(ordered_chars)
rotation = 270
if matrix[0] < 0 and matrix[3] < 0:
rotation = 180
elif matrix[1] > 0:
rotation = 90
word = {
"text": "".join(
self.expansions.get(c["text"], c["text"]) for c in ordered_chars
),
"x0": x0,
"x1": x1,
"top": top,
"doctop": top + doctop_adj,
"bottom": bottom,
"upright": upright,
"direction": direction,
"rotation": rotation,
}
for key in self.extra_attrs:
word[key] = ordered_chars[0][key]
return word
def char_begins_new_word(
self,
prev_char,
curr_char,
) -> bool:
"""This method takes several factors into account to determine if
`curr_char` represents the beginning of a new word:
- Whether the text is "upright" (i.e., non-rotated)
- Whether the user has specified that horizontal text runs
left-to-right (default) or right-to-left, as represented by
self.horizontal_ltr
- Whether the user has specified that vertical text the text runs
top-to-bottom (default) or bottom-to-top, as represented by
self.vertical_ttb
- The x0, top, x1, and bottom attributes of prev_char and
curr_char
- The self.x_tolerance and self.y_tolerance settings. Note: In
this case, x/y refer to those directions for non-rotated text.
For vertical text, they are flipped. A more accurate terminology
might be "*intra*line character distance tolerance" and
"*inter*line character distance tolerance"
An important note: The *intra*line distance is measured from the
*end* of the previous character to the *beginning* of the current
character, while the *inter*line distance is measured from the
*top* of the previous character to the *top* of the next
character. The reasons for this are partly repository-historical,
and partly logical, as successive text lines' bounding boxes often
overlap slightly (and we don't want that overlap to be interpreted
as the two lines being the same line).
The upright-ness of the character determines the attributes to
compare, while horizontal_ltr/vertical_ttb determine the direction
of the comparison.
"""
# Note: Due to the grouping step earlier in the process,
# curr_char["upright"] will always equal prev_char["upright"].
if curr_char["upright"]:
x = self.x_tolerance
y = self.y_tolerance
ay = prev_char["top"]
cy = curr_char["top"]
if self.horizontal_ltr:
ax = prev_char["x0"]
bx = prev_char["x1"]
cx = curr_char["x0"]
else:
ax = -prev_char["x1"]
bx = -prev_char["x0"]
cx = -curr_char["x1"]
else:
x = self.y_tolerance
y = self.x_tolerance
ay = prev_char["x0"]
cy = curr_char["x0"]
if self.vertical_ttb:
ax = prev_char["top"]
bx = prev_char["bottom"]
cx = curr_char["top"]
else:
ax = -prev_char["bottom"]
bx = -prev_char["top"]
cx = -curr_char["bottom"]
return bool(
# Intraline test
(cx < ax)
or (cx > bx + x)
# Interline test
or (cy > ay + y)
)
def iter_chars_to_words(self, ordered_chars):
current_word: list = []
def start_next_word(new_char=None):
nonlocal current_word
if current_word:
yield current_word
current_word = [] if new_char is None else [new_char]
for char in ordered_chars:
text = char["text"]
if not self.keep_blank_chars and text.isspace():
yield from start_next_word(None)
elif text in self.split_at_punctuation:
yield from start_next_word(char)
yield from start_next_word(None)
elif current_word and self.char_begins_new_word(current_word[-1], char):
yield from start_next_word(char)
else:
current_word.append(char)
# Finally, after all chars processed
if current_word:
yield current_word
def iter_sort_chars(self, chars):
def upright_key(x) -> int:
return -int(x["upright"])
for upright_cluster in cluster_objects(list(chars), upright_key, 0):
upright = upright_cluster[0]["upright"]
cluster_key = "doctop" if upright else "x0"
# Cluster by line
subclusters = cluster_objects(
upright_cluster, itemgetter(cluster_key), self.y_tolerance
)
for sc in subclusters:
# Sort within line
sort_key = "x0" if upright else "doctop"
to_yield = sorted(sc, key=itemgetter(sort_key))
# Reverse order if necessary
if not (self.horizontal_ltr if upright else self.vertical_ttb):
yield from reversed(to_yield)
else:
yield from to_yield
def iter_extract_tuples(self, chars):
ordered_chars = chars if self.use_text_flow else self.iter_sort_chars(chars)
grouping_key = itemgetter("upright", *self.extra_attrs)
grouped_chars = itertools.groupby(ordered_chars, grouping_key)
for keyvals, char_group in grouped_chars:
for word_chars in self.iter_chars_to_words(char_group):
yield (self.merge_chars(word_chars), word_chars)
def extract_wordmap(self, chars) -> WordMap:
return WordMap(list(self.iter_extract_tuples(chars)))
def extract_words(self, chars: list) -> list:
words = list(word for word, word_chars in self.iter_extract_tuples(chars))
return words
def extract_words(chars: list, **kwargs) -> list:
return WordExtractor(**kwargs).extract_words(chars)
TEXTMAP_KWARGS = inspect.signature(WordMap.to_textmap).parameters.keys()
WORD_EXTRACTOR_KWARGS = inspect.signature(WordExtractor).parameters.keys()
def chars_to_textmap(chars: list, **kwargs) -> TextMap:
kwargs.update({"presorted": True})
extractor = WordExtractor(
**{k: kwargs[k] for k in WORD_EXTRACTOR_KWARGS if k in kwargs}
)
wordmap = extractor.extract_wordmap(chars)
textmap = wordmap.to_textmap(
**{k: kwargs[k] for k in TEXTMAP_KWARGS if k in kwargs}
)
return textmap
def extract_text(chars: list, **kwargs) -> str:
chars = to_list(chars)
if len(chars) == 0:
return ""
if kwargs.get("layout"):
return chars_to_textmap(chars, **kwargs).as_string
else:
y_tolerance = kwargs.get("y_tolerance", DEFAULT_Y_TOLERANCE)
extractor = WordExtractor(
**{k: kwargs[k] for k in WORD_EXTRACTOR_KWARGS if k in kwargs}
)
words = extractor.extract_words(chars)
if words:
rotation = words[0]["rotation"] # rotation cannot change within a cell
else:
rotation = 0
if rotation == 90:
words.sort(key=lambda w: (w["x1"], -w["top"]))
lines = " ".join([w["text"] for w in words])
elif rotation == 270:
words.sort(key=lambda w: (-w["x1"], w["top"]))
lines = " ".join([w["text"] for w in words])
else:
lines = cluster_objects(words, itemgetter("doctop"), y_tolerance)
lines = "\n".join(" ".join(word["text"] for word in line) for line in lines)
if rotation == 180: # needs extra treatment
lines = "".join([(c if c != "\n" else " ") for c in reversed(lines)])
return lines
def collate_line(
line_chars: list,
tolerance=DEFAULT_X_TOLERANCE,
) -> str:
coll = ""
last_x1 = None
for char in sorted(line_chars, key=itemgetter("x0")):
if (last_x1 is not None) and (char["x0"] > (last_x1 + tolerance)):
coll += " "
last_x1 = char["x1"]
coll += char["text"]
return coll
def dedupe_chars(chars: list, tolerance=1) -> list:
"""
Removes duplicate chars — those sharing the same text, fontname, size,
and positioning (within `tolerance`) as other characters in the set.
"""
key = itemgetter("fontname", "size", "upright", "text")
pos_key = itemgetter("doctop", "x0")
def yield_unique_chars(chars: list):
sorted_chars = sorted(chars, key=key)
for grp, grp_chars in itertools.groupby(sorted_chars, key=key):
for y_cluster in cluster_objects(
list(grp_chars), itemgetter("doctop"), tolerance
):
for x_cluster in cluster_objects(
y_cluster, itemgetter("x0"), tolerance
):
yield sorted(x_cluster, key=pos_key)[0]
deduped = yield_unique_chars(chars)
return sorted(deduped, key=chars.index)
def line_to_edge(line):
edge = dict(line)
edge["orientation"] = "h" if (line["top"] == line["bottom"]) else "v"
return edge
def rect_to_edges(rect) -> list:
top, bottom, left, right = [dict(rect) for x in range(4)]
top.update(
{
"object_type": "rect_edge",
"height": 0,
"y0": rect["y1"],
"bottom": rect["top"],
"orientation": "h",
}
)
bottom.update(
{
"object_type": "rect_edge",
"height": 0,
"y1": rect["y0"],
"top": rect["top"] + rect["height"],
"doctop": rect["doctop"] + rect["height"],
"orientation": "h",
}
)
left.update(
{
"object_type": "rect_edge",
"width": 0,
"x1": rect["x0"],
"orientation": "v",
}
)
right.update(
{
"object_type": "rect_edge",
"width": 0,
"x0": rect["x1"],
"orientation": "v",
}
)
return [top, bottom, left, right]
def curve_to_edges(curve) -> list:
point_pairs = zip(curve["pts"], curve["pts"][1:])
return [
{
"object_type": "curve_edge",
"x0": min(p0[0], p1[0]),
"x1": max(p0[0], p1[0]),
"top": min(p0[1], p1[1]),
"doctop": min(p0[1], p1[1]) + (curve["doctop"] - curve["top"]),
"bottom": max(p0[1], p1[1]),
"width": abs(p0[0] - p1[0]),
"height": abs(p0[1] - p1[1]),
"orientation": "v" if p0[0] == p1[0] else ("h" if p0[1] == p1[1] else None),
}
for p0, p1 in point_pairs
]
def obj_to_edges(obj) -> list:
t = obj["object_type"]
if "_edge" in t:
return [obj]
elif t == "line":
return [line_to_edge(obj)]
else:
return {"rect": rect_to_edges, "curve": curve_to_edges}[t](obj)
def filter_edges(
edges,
orientation=None,
edge_type=None,
min_length=1,
) -> list:
if orientation not in ("v", "h", None):
raise ValueError("Orientation must be 'v' or 'h'")
def test(e) -> bool:
dim = "height" if e["orientation"] == "v" else "width"
et_correct = e["object_type"] == edge_type if edge_type is not None else True
orient_correct = orientation is None or e["orientation"] == orientation
return bool(et_correct and orient_correct and (e[dim] >= min_length))
return list(filter(test, edges))
def cluster_list(xs, tolerance=0) -> list:
if tolerance == 0:
return [[x] for x in sorted(xs)]
if len(xs) < 2:
return [[x] for x in sorted(xs)]
groups = []
xs = list(sorted(xs))
current_group = [xs[0]]
last = xs[0]
for x in xs[1:]:
if x <= (last + tolerance):
current_group.append(x)
else:
groups.append(current_group)
current_group = [x]
last = x
groups.append(current_group)
return groups
def make_cluster_dict(values, tolerance) -> dict:
clusters = cluster_list(list(set(values)), tolerance)
nested_tuples = [
[(val, i) for val in value_cluster] for i, value_cluster in enumerate(clusters)
]
return dict(itertools.chain(*nested_tuples))
def cluster_objects(xs, key_fn, tolerance) -> list:
if not callable(key_fn):
key_fn = itemgetter(key_fn)
values = map(key_fn, xs)
cluster_dict = make_cluster_dict(values, tolerance)
get_0, get_1 = itemgetter(0), itemgetter(1)
cluster_tuples = sorted(((x, cluster_dict.get(key_fn(x))) for x in xs), key=get_1)
grouped = itertools.groupby(cluster_tuples, key=get_1)
return [list(map(get_0, v)) for k, v in grouped]
def move_object(obj, axis: str, value):
assert axis in ("h", "v")
if axis == "h":
new_items = [
("x0", obj["x0"] + value),
("x1", obj["x1"] + value),
]
if axis == "v":
new_items = [
("top", obj["top"] + value),
("bottom", obj["bottom"] + value),
]
if "doctop" in obj:
new_items += [("doctop", obj["doctop"] + value)]
if "y0" in obj:
new_items += [
("y0", obj["y0"] - value),
("y1", obj["y1"] - value),
]
return obj.__class__(tuple(obj.items()) + tuple(new_items))
def snap_objects(objs, attr: str, tolerance) -> list:
axis = {"x0": "h", "x1": "h", "top": "v", "bottom": "v"}[attr]
list_objs = list(objs)
clusters = cluster_objects(list_objs, itemgetter(attr), tolerance)
avgs = [sum(map(itemgetter(attr), cluster)) / len(cluster) for cluster in clusters]
snapped_clusters = [
[move_object(obj, axis, avg - obj[attr]) for obj in cluster]
for cluster, avg in zip(clusters, avgs)
]
return list(itertools.chain(*snapped_clusters))
def snap_edges(
edges,
x_tolerance=DEFAULT_SNAP_TOLERANCE,
y_tolerance=DEFAULT_SNAP_TOLERANCE,
):
"""
Given a list of edges, snap any within `tolerance` pixels of one another
to their positional average.
"""
by_orientation = {"v": [], "h": []}
for e in edges:
by_orientation[e["orientation"]].append(e)
snapped_v = snap_objects(by_orientation["v"], "x0", x_tolerance)
snapped_h = snap_objects(by_orientation["h"], "top", y_tolerance)
return snapped_v + snapped_h
def resize_object(obj, key: str, value):
assert key in ("x0", "x1", "top", "bottom")
old_value = obj[key]
diff = value - old_value
new_items = [
(key, value),
]
if key == "x0":
assert value <= obj["x1"]
new_items.append(("width", obj["x1"] - value))
elif key == "x1":
assert value >= obj["x0"]
new_items.append(("width", value - obj["x0"]))
elif key == "top":
assert value <= obj["bottom"]
new_items.append(("doctop", obj["doctop"] + diff))
new_items.append(("height", obj["height"] - diff))
if "y1" in obj:
new_items.append(("y1", obj["y1"] - diff))
elif key == "bottom":
assert value >= obj["top"]
new_items.append(("height", obj["height"] + diff))
if "y0" in obj:
new_items.append(("y0", obj["y0"] - diff))
return obj.__class__(tuple(obj.items()) + tuple(new_items))
def join_edge_group(edges, orientation: str, tolerance=DEFAULT_JOIN_TOLERANCE):
"""
Given a list of edges along the same infinite line, join those that
are within `tolerance` pixels of one another.
"""
if orientation == "h":
min_prop, max_prop = "x0", "x1"
elif orientation == "v":
min_prop, max_prop = "top", "bottom"
else:
raise ValueError("Orientation must be 'v' or 'h'")
sorted_edges = list(sorted(edges, key=itemgetter(min_prop)))
joined = [sorted_edges[0]]
for e in sorted_edges[1:]:
last = joined[-1]
if e[min_prop] <= (last[max_prop] + tolerance):
if e[max_prop] > last[max_prop]:
# Extend current edge to new extremity
joined[-1] = resize_object(last, max_prop, e[max_prop])
else:
# Edge is separate from previous edges
joined.append(e)
return joined
def merge_edges(
edges,
snap_x_tolerance,
snap_y_tolerance,
join_x_tolerance,
join_y_tolerance,
):
"""
Using the `snap_edges` and `join_edge_group` methods above,
merge a list of edges into a more "seamless" list.
"""
def get_group(edge):
if edge["orientation"] == "h":
return ("h", edge["top"])
else:
return ("v", edge["x0"])
if snap_x_tolerance > 0 or snap_y_tolerance > 0:
edges = snap_edges(edges, snap_x_tolerance, snap_y_tolerance)
_sorted = sorted(edges, key=get_group)
edge_groups = itertools.groupby(_sorted, key=get_group)
edge_gen = (
join_edge_group(
items, k[0], (join_x_tolerance if k[0] == "h" else join_y_tolerance)
)
for k, items in edge_groups
)
edges = list(itertools.chain(*edge_gen))
return edges
def bbox_to_rect(bbox) -> dict:
"""
Return the rectangle (i.e a dict with keys "x0", "top", "x1",
"bottom") for an object.
"""
return {"x0": bbox[0], "top": bbox[1], "x1": bbox[2], "bottom": bbox[3]}
def objects_to_rect(objects) -> dict:
"""
Given an iterable of objects, return the smallest rectangle (i.e. a
dict with "x0", "top", "x1", and "bottom" keys) that contains them
all.
"""
return bbox_to_rect(objects_to_bbox(objects))
def merge_bboxes(bboxes):
"""
Given an iterable of bounding boxes, return the smallest bounding box
that contains them all.
"""
x0, top, x1, bottom = zip(*bboxes)
return (min(x0), min(top), max(x1), max(bottom))
def objects_to_bbox(objects):
"""
Given an iterable of objects, return the smallest bounding box that
contains them all.
"""
return merge_bboxes(map(bbox_getter, objects))
def words_to_edges_h(words, word_threshold: int = DEFAULT_MIN_WORDS_HORIZONTAL):
"""
Find (imaginary) horizontal lines that connect the tops
of at least `word_threshold` words.
"""
by_top = cluster_objects(words, itemgetter("top"), 1)
large_clusters = filter(lambda x: len(x) >= word_threshold, by_top)
rects = list(map(objects_to_rect, large_clusters))
if len(rects) == 0:
return []
min_x0 = min(map(itemgetter("x0"), rects))
max_x1 = max(map(itemgetter("x1"), rects))
edges = []
for r in rects:
edges += [
# Top of text
{
"x0": min_x0,
"x1": max_x1,
"top": r["top"],
"bottom": r["top"],
"width": max_x1 - min_x0,
"orientation": "h",
},
# For each detected row, we also add the 'bottom' line. This will
# generate extra edges, (some will be redundant with the next row
# 'top' line), but this catches the last row of every table.
{
"x0": min_x0,
"x1": max_x1,
"top": r["bottom"],
"bottom": r["bottom"],
"width": max_x1 - min_x0,
"orientation": "h",
},
]
return edges
def get_bbox_overlap(a, b):
a_left, a_top, a_right, a_bottom = a
b_left, b_top, b_right, b_bottom = b
o_left = max(a_left, b_left)
o_right = min(a_right, b_right)
o_bottom = min(a_bottom, b_bottom)
o_top = max(a_top, b_top)
o_width = o_right - o_left
o_height = o_bottom - o_top
if o_height >= 0 and o_width >= 0 and o_height + o_width > 0:
return (o_left, o_top, o_right, o_bottom)
else:
return None
def words_to_edges_v(words, word_threshold: int = DEFAULT_MIN_WORDS_VERTICAL):
"""
Find (imaginary) vertical lines that connect the left, right, or
center of at least `word_threshold` words.
"""
# Find words that share the same left, right, or centerpoints
by_x0 = cluster_objects(words, itemgetter("x0"), 1)
by_x1 = cluster_objects(words, itemgetter("x1"), 1)
def get_center(word):
return float(word["x0"] + word["x1"]) / 2
by_center = cluster_objects(words, get_center, 1)
clusters = by_x0 + by_x1 + by_center
# Find the points that align with the most words
sorted_clusters = sorted(clusters, key=lambda x: -len(x))
large_clusters = filter(lambda x: len(x) >= word_threshold, sorted_clusters)
# For each of those points, find the bboxes fitting all matching words
bboxes = list(map(objects_to_bbox, large_clusters))
# Iterate through those bboxes, condensing overlapping bboxes
condensed_bboxes = []
for bbox in bboxes:
overlap = any(get_bbox_overlap(bbox, c) for c in condensed_bboxes)
if not overlap:
condensed_bboxes.append(bbox)
if not condensed_bboxes:
return []
condensed_rects = map(bbox_to_rect, condensed_bboxes)
sorted_rects = list(sorted(condensed_rects, key=itemgetter("x0")))
max_x1 = max(map(itemgetter("x1"), sorted_rects))
min_top = min(map(itemgetter("top"), sorted_rects))
max_bottom = max(map(itemgetter("bottom"), sorted_rects))
return [
{
"x0": b["x0"],
"x1": b["x0"],
"top": min_top,
"bottom": max_bottom,
"height": max_bottom - min_top,
"orientation": "v",
}
for b in sorted_rects
] + [
{
"x0": max_x1,
"x1": max_x1,
"top": min_top,
"bottom": max_bottom,
"height": max_bottom - min_top,
"orientation": "v",
}
]
def edges_to_intersections(edges, x_tolerance=1, y_tolerance=1) -> dict:
"""
Given a list of edges, return the points at which they intersect
within `tolerance` pixels.
"""
intersections = {}
v_edges, h_edges = [
list(filter(lambda x: x["orientation"] == o, edges)) for o in ("v", "h")
]
for v in sorted(v_edges, key=itemgetter("x0", "top")):
for h in sorted(h_edges, key=itemgetter("top", "x0")):
if (
(v["top"] <= (h["top"] + y_tolerance))
and (v["bottom"] >= (h["top"] - y_tolerance))
and (v["x0"] >= (h["x0"] - x_tolerance))
and (v["x0"] <= (h["x1"] + x_tolerance))
):
vertex = (v["x0"], h["top"])
if vertex not in intersections:
intersections[vertex] = {"v": [], "h": []}
intersections[vertex]["v"].append(v)
intersections[vertex]["h"].append(h)
return intersections
def obj_to_bbox(obj):
"""
Return the bounding box for an object.
"""
return bbox_getter(obj)
def intersections_to_cells(intersections):
"""
Given a list of points (`intersections`), return all rectangular "cells"
that those points describe.
`intersections` should be a dictionary with (x0, top) tuples as keys,
and a list of edge objects as values. The edge objects should correspond
to the edges that touch the intersection.
"""
def edge_connects(p1, p2) -> bool:
def edges_to_set(edges):
return set(map(obj_to_bbox, edges))
if p1[0] == p2[0]:
common = edges_to_set(intersections[p1]["v"]).intersection(
edges_to_set(intersections[p2]["v"])
)
if len(common):
return True
if p1[1] == p2[1]:
common = edges_to_set(intersections[p1]["h"]).intersection(
edges_to_set(intersections[p2]["h"])
)
if len(common):
return True
return False
points = list(sorted(intersections.keys()))
n_points = len(points)
def find_smallest_cell(points, i: int):
if i == n_points - 1:
return None
pt = points[i]
rest = points[i + 1 :]
# Get all the points directly below and directly right
below = [x for x in rest if x[0] == pt[0]]
right = [x for x in rest if x[1] == pt[1]]
for below_pt in below:
if not edge_connects(pt, below_pt):
continue
for right_pt in right:
if not edge_connects(pt, right_pt):
continue
bottom_right = (right_pt[0], below_pt[1])
if (
(bottom_right in intersections)
and edge_connects(bottom_right, right_pt)
and edge_connects(bottom_right, below_pt)
):
return (pt[0], pt[1], bottom_right[0], bottom_right[1])
return None
cell_gen = (find_smallest_cell(points, i) for i in range(len(points)))
return list(filter(None, cell_gen))
def cells_to_tables(page, cells) -> list:
"""
Given a list of bounding boxes (`cells`), return a list of tables that
hold those cells most simply (and contiguously).
"""
def bbox_to_corners(bbox) -> tuple:
x0, top, x1, bottom = bbox
return ((x0, top), (x0, bottom), (x1, top), (x1, bottom))
remaining_cells = list(cells)
# Iterate through the cells found above, and assign them
# to contiguous tables
current_corners = set()
current_cells = []
tables = []
while len(remaining_cells):
initial_cell_count = len(current_cells)
for cell in list(remaining_cells):
cell_corners = bbox_to_corners(cell)
# If we're just starting a table ...
if len(current_cells) == 0:
# ... immediately assign it to the empty group
current_corners |= set(cell_corners)
current_cells.append(cell)
remaining_cells.remove(cell)
else:
# How many corners does this table share with the current group?
corner_count = sum(c in current_corners for c in cell_corners)
# If touching on at least one corner...
if corner_count > 0:
# ... assign it to the current group
current_corners |= set(cell_corners)
current_cells.append(cell)
remaining_cells.remove(cell)
# If this iteration did not find any more cells to append...
if len(current_cells) == initial_cell_count:
# ... start a new cell group
tables.append(list(current_cells))
current_corners.clear()
current_cells.clear()
# Once we have exhausting the list of cells ...
# ... and we have a cell group that has not been stored
if len(current_cells):
# ... store it.
tables.append(list(current_cells))
# PyMuPDF modification:
# Remove tables without text or having only 1 column.
# The text probe scans the current call's CHARS: the module-global
# TEXTPAGE it used to pass to page.get_textbox() was never assigned, so
# every probe built and walked a fresh full-page TextPage per candidate.
# The overlap test mirrors JM_rects_overlap() (strict), and a candidate
# counts as texty iff it overlaps a non-whitespace character.
text_bboxes = None
def has_text(r):
nonlocal text_bboxes
if text_bboxes is None: # built once, only if a candidate survives
text_bboxes = [
(c["x0"], c["top"], c["x1"], c["bottom"])
for c in CHARS
if c["text"] not in white_spaces
]
rx0, rtop, rx1, rbottom = r
for x0, top, x1, bottom in text_bboxes:
if x0 < rx1 and x1 > rx0 and top < rbottom and bottom > rtop:
return True
return False
for i in range(len(tables) - 1, -1, -1):
r = pymupdf.EMPTY_RECT()
x1_vals = set()
x0_vals = set()
for c in tables[i]:
r |= c
x1_vals.add(c[2])
x0_vals.add(c[0])
if (
len(x1_vals) < 2
or len(x0_vals) < 2
or not has_text(r)
):
del tables[i]
# Sort the tables top-to-bottom-left-to-right based on the value of the
# topmost-and-then-leftmost coordinate of a table.
_sorted = sorted(tables, key=lambda t: min((c[1], c[0]) for c in t))
return _sorted
class CellGroup:
def __init__(self, cells):
self.cells = cells
self.bbox = (
min(map(itemgetter(0), filter(None, cells))),
min(map(itemgetter(1), filter(None, cells))),
max(map(itemgetter(2), filter(None, cells))),
max(map(itemgetter(3), filter(None, cells))),
)
class TableRow(CellGroup):
pass
class TableHeader:
"""PyMuPDF extension containing the identified table header."""
def __init__(self, bbox, cells, names, above):
self.bbox = bbox
self.cells = cells
self.names = names
self.external = above
class Table:
def __init__(self, page, cells, bbox=None):
self.page = page
self.textpage = None
self._chars = None
self.cells = cells
# Explicit reported-bbox override; None means bbox is computed from
# cells. Set only for a union grid-ref table, whose reported region
# (its layout box) is decoupled from its replacement cell grid.
self._bbox = bbox
self.header = self._get_header() # PyMuPDF extension
# Filled by find_tables(refine=True): placements is a row-major grid of
# tagged SpanCell colspan/rowspan placements (None otherwise); header_rows
# is the leading header-row count, section_rows the section-label rows.
self.placements = None
self.header_rows = 0
self.section_rows = ()
@property
def bbox(self):
if self._bbox is not None:
return tuple(self._bbox)
c = self.cells
return (
min(map(itemgetter(0), c)),
min(map(itemgetter(1), c)),
max(map(itemgetter(2), c)),
max(map(itemgetter(3), c)),
)
@property
def rows(self) -> list:
_sorted = sorted(self.cells, key=itemgetter(1, 0))
xs = list(sorted(set(map(itemgetter(0), self.cells))))
rows = []
for y, row_cells in itertools.groupby(_sorted, itemgetter(1)):
xdict = {cell[0]: cell for cell in row_cells}
row = TableRow([xdict.get(x) for x in xs])
rows.append(row)
return rows
@property
def row_count(self) -> int: # PyMuPDF extension
return len(self.rows)
@property
def col_count(self) -> int: # PyMuPDF extension
return max([len(r.cells) for r in self.rows])
def extract(self, **kwargs) -> list:
chars = self._chars if self._chars is not None else CHARS
table_arr = []
def char_in_bbox(char, bbox) -> bool:
v_mid = (char["top"] + char["bottom"]) / 2
h_mid = (char["x0"] + char["x1"]) / 2
x0, top, x1, bottom = bbox
return bool(
(h_mid >= x0) and (h_mid < x1) and (v_mid >= top) and (v_mid < bottom)
)
for row in self.rows:
arr = []
row_chars = [char for char in chars if char_in_bbox(char, row.bbox)]
for cell in row.cells:
if cell is None:
cell_text = None
else:
cell_chars = [
char for char in row_chars if char_in_bbox(char, cell)
]
if len(cell_chars):
kwargs["x_shift"] = cell[0]
kwargs["y_shift"] = cell[1]
if "layout" in kwargs:
kwargs["layout_width"] = cell[2] - cell[0]
kwargs["layout_height"] = cell[3] - cell[1]
cell_text = extract_text(cell_chars, **kwargs)
else:
cell_text = ""
arr.append(cell_text)
table_arr.append(arr)
return table_arr
def to_markdown(self, clean=False, fill_empty=True):
"""Output table content as a string in Github-markdown format.
If "clean" then markdown syntax is removed from cell content.
If "fill_empty" then cell content None is replaced by the values
above (columns) or left (rows) in an effort to approximate row and
columns spans.
"""
output = "|"
rows = self.row_count
cols = self.col_count
# cell coordinates
cell_boxes = [[c for c in r.cells] for r in self.rows]
# cell text strings
cells = [[None for i in range(cols)] for j in range(rows)]
for i, row in enumerate(cell_boxes):
for j, cell in enumerate(row):
if cell is not None:
cells[i][j] = extract_cells(
self.textpage, cell_boxes[i][j], markdown=True
)
if fill_empty: # fill "None" cells where possible
# for rows, copy content from left to right
for j in range(rows):
for i in range(cols - 1):
if cells[j][i + 1] is None:
cells[j][i + 1] = cells[j][i]
# for columns, copy top to bottom
for i in range(cols):
for j in range(rows - 1):
if cells[j + 1][i] is None:
cells[j + 1][i] = cells[j][i]
# generate header string and MD separator
for i, name in enumerate(self.header.names):
if not name: # generate a name if empty
name = f"Col{i+1}"
name = name.replace("\n", "<br>") # use HTML line breaks
if clean: # remove sensitive syntax
name = html.escape(name.replace("-", "-"))
output += name + "|"
output += "\n"
# insert GitHub header line separator
output += "|" + "|".join("---" for i in range(self.col_count)) + "|\n"
# skip first row in details if header is part of the table
j = 0 if self.header.external else 1
# iterate over detail rows
for row in cells[j:]:
line = "|"
for i, cell in enumerate(row):
# replace None cells with empty string
# use HTML line break tag
if cell is None:
cell = ""
if clean: # remove sensitive syntax
cell = html.escape(cell.replace("-", "-"))
line += cell + "|"
line += "\n"
output += line
return output + "\n"
def to_html(self):
"""Output the table as an HTML ``<table>`` string.
With span placements (from ``find_tables(refine=True)``) each
placement's colspan/rowspan and th/td tag are honoured and a
section-label row collapses to one spanning ``<th>``. Otherwise a flat
td-only table is built from :meth:`extract`.
"""
if self.placements is not None:
return render_table_html(self.placements, self.section_rows)
# No placements: flat td-only grid from extract().
rows = [
[
SpanCell(bbox=None, text=(cell or ""), colspan=1, rowspan=1)
for cell in row
]
for row in self.extract()
]
return render_table_html(rows)
def to_pandas(self, **kwargs):
"""Return a pandas DataFrame version of the table."""
try:
import pandas as pd
except ModuleNotFoundError:
pymupdf.message("Package 'pandas' is not installed")
raise
pd_dict = {}
extract = self.extract()
hdr = self.header
names = self.header.names
hdr_len = len(names)
# ensure uniqueness of column names
for i in range(hdr_len):
name = names[i]
if not name:
names[i] = f"Col{i}"
if hdr_len != len(set(names)):
for i in range(hdr_len):
name = names[i]
if name != f"Col{i}":
names[i] = f"{i}-{name}"
if not hdr.external: # header is part of 'extract'
extract = extract[1:]
for i in range(hdr_len):
key = names[i]
value = []
for j in range(len(extract)):
value.append(extract[j][i])
pd_dict[key] = value
return pd.DataFrame(pd_dict)
def _get_header(self, y_tolerance=3):
"""Identify the table header.
*** PyMuPDF extension. ***
Starting from the first line above the table upwards, check if it
qualifies to be part of the table header.
Criteria include:
* A one-line table never has an extra header.
* Column borders must not intersect any word. If this happens, all
text of this line and above of it is ignored.
* No excess inter-line distance: If a line further up has a distance
of more than 1.5 times of its font size, it will be ignored and
all lines above of it.
* Must have same text properties.
* Starting with the top table line, a bold text property cannot change
back to non-bold.
If not all criteria are met (or there is no text above the table),
the first table row is assumed to be the header.
"""
page = self.page
y_delta = y_tolerance
def top_row_bg_color(self):
"""
Compare top row background color with color of same-sized bbox
above. If different, return True indicating that the original
table top row is already the header.
"""
bbox0 = pymupdf.Rect(self.rows[0].bbox)
bboxt = bbox0 + (0, -bbox0.height, 0, -bbox0.height) # area above
top_color0 = page.get_pixmap(clip=bbox0).color_topusage()[1]
top_colort = page.get_pixmap(clip=bboxt).color_topusage()[1]
if top_color0 != top_colort:
return True # top row is header
return False
def row_has_bold(bbox):
"""Check if a row contains some bold text.
If e.g. true for the top row, then it will be used as (internal)
column header row if any of the following is true:
* the previous (above) text line has no bold span
* the second table row text has no bold span
Returns True if any spans are bold else False.
"""
return any(
c["bold"]
for c in CHARS
if rect_in_rect((c["x0"], c["y0"], c["x1"], c["y1"]), bbox)
)
try:
row = self.rows[0]
cells = row.cells
bbox = pymupdf.Rect(row.bbox)
except IndexError: # this table has no rows
return None
# return this if we determine that the top row is the header
header_top_row = TableHeader(bbox, cells, self.extract()[0], False)
# 1-line tables have no extra header
if len(self.rows) < 2:
return header_top_row
# 1-column tables have no extra header
if len(cells) < 2:
return header_top_row
# assume top row is the header if second row is empty
row2 = self.rows[1] # second row
if all(c is None for c in row2.cells): # no valid cell bboxes in row2
return header_top_row
# Special check: is top row bold?
top_row_bold = row_has_bold(bbox)
# assume top row is header if it is bold and any cell
# of 2nd row is non-bold
if top_row_bold and not row_has_bold(row2.bbox):
return header_top_row
if top_row_bg_color(self):
# if area above top row has a different background color,
# then top row is already the header
return header_top_row
# column coordinates (x1 values) in top row
col_x = [c[2] if c is not None else None for c in cells[:-1]]
# clip = page area above the table
# We will inspect this area for text qualifying as column header.
clip = +bbox # take row 0 bbox
clip.y0 = 0 # start at top of page
clip.y1 = bbox.y0 # end at top of table
blocks = page.get_text("dict", clip=clip, flags=pymupdf.TEXTFLAGS_TEXT)[
"blocks"
]
# non-empty, non-superscript spans above table, sorted descending by y1
spans = sorted(
[
s
for b in blocks
for l in b["lines"]
for s in l["spans"]
if not (
white_spaces.issuperset(s["text"])
or s["flags"] & pymupdf.TEXT_FONT_SUPERSCRIPT
)
],
key=lambda s: s["bbox"][3],
reverse=True,
)
select = [] # y1 coordinates above, sorted descending
line_heights = [] # line heights above, sorted descending
line_bolds = [] # bold indicator per line above, same sorting
# walk through the spans and fill above 3 lists
for i in range(len(spans)):
s = spans[i]
y1 = s["bbox"][3] # span bottom
h = y1 - s["bbox"][1] # span bbox height
bold = s["flags"] & pymupdf.TEXT_FONT_BOLD
# use first item to start the lists
if i == 0:
select.append(y1)
line_heights.append(h)
line_bolds.append(bold)
continue
# get previous items from the 3 lists
y0 = select[-1]
h0 = line_heights[-1]
bold0 = line_bolds[-1]
if bold0 and not bold:
break # stop if switching from bold to non-bold
# if fitting in height of previous span, modify bbox
if y0 - y1 <= y_delta or abs((y0 - h0) - s["bbox"][1]) <= y_delta:
s["bbox"] = (s["bbox"][0], y0 - h0, s["bbox"][2], y0)
spans[i] = s
if bold:
line_bolds[-1] = bold
continue
elif y0 - y1 > 1.5 * h0:
break # stop if distance to previous line too large
select.append(y1)
line_heights.append(h)
line_bolds.append(bold)
if select == []: # nothing above the table?
return header_top_row
select = select[:5] # accept up to 5 lines for an external header
# assume top row as header if text above is too far away
if bbox.y0 - select[0] >= line_heights[0]:
return header_top_row
# accept top row as header if bold, but line above is not
if top_row_bold and not line_bolds[0]:
return header_top_row
if spans == []: # nothing left above the table, return top row
return header_top_row
# re-compute clip above table
nclip = pymupdf.EMPTY_RECT()
for s in [s for s in spans if s["bbox"][3] >= select[-1]]:
nclip |= s["bbox"]
if not nclip.is_empty:
clip = nclip
clip.y1 = bbox.y0 # make sure we still include every word above
# Confirm that no word in clip is intersecting a column separator
word_rects = [pymupdf.Rect(w[:4]) for w in page.get_text("words", clip=clip)]
word_tops = sorted(list(set([r[1] for r in word_rects])), reverse=True)
select = []
# exclude lines with words that intersect a column border
for top in word_tops:
intersecting = [
(x, r)
for x in col_x
if x is not None
for r in word_rects
if r[1] == top and r[0] < x and r[2] > x
]
if intersecting == []:
select.append(top)
else: # detected a word crossing a column border
break
if select == []: # nothing left over: return first row
return header_top_row
hdr_bbox = +clip # compute the header cells
hdr_bbox.y0 = select[-1] # hdr_bbox top is smallest top coord of words
hdr_cells = [
(c[0], hdr_bbox.y0, c[2], hdr_bbox.y1) if c is not None else None
for c in cells
]
# adjust left/right of header bbox
hdr_bbox.x0 = self.bbox[0]
hdr_bbox.x1 = self.bbox[2]
# column names: no line breaks, no excess spaces
hdr_names = [
(
page.get_textbox(c).replace("\n", " ").replace(" ", " ").strip()
if c is not None
else ""
)
for c in hdr_cells
]
return TableHeader(tuple(hdr_bbox), hdr_cells, hdr_names, True)
@dataclass
class TableSettings:
vertical_strategy: str = "lines"
horizontal_strategy: str = "lines"
explicit_vertical_lines: list = None
explicit_horizontal_lines: list = None
snap_tolerance: float = DEFAULT_SNAP_TOLERANCE
snap_x_tolerance: float = UNSET
snap_y_tolerance: float = UNSET
join_tolerance: float = DEFAULT_JOIN_TOLERANCE
join_x_tolerance: float = UNSET
join_y_tolerance: float = UNSET
edge_min_length: float = 3
min_words_vertical: float = DEFAULT_MIN_WORDS_VERTICAL
min_words_horizontal: float = DEFAULT_MIN_WORDS_HORIZONTAL
intersection_tolerance: float = 3
intersection_x_tolerance: float = UNSET
intersection_y_tolerance: float = UNSET
text_settings: dict = None
def __post_init__(self) -> "TableSettings":
"""Clean up user-provided table settings.
Validates that the table settings provided consists of acceptable values and
returns a cleaned up version. The cleaned up version fills out the missing
values with the default values in the provided settings.
TODO: Can be further used to validate that the values are of the correct
type. For example, raising a value error when a non-boolean input is
provided for the key ``keep_blank_chars``.
:param table_settings: User-provided table settings.
:returns: A cleaned up version of the user-provided table settings.
:raises ValueError: When an unrecognised key is provided.
"""
for setting in NON_NEGATIVE_SETTINGS:
if (getattr(self, setting) or 0) < 0:
raise ValueError(f"Table setting '{setting}' cannot be negative")
for orientation in ["horizontal", "vertical"]:
strategy = getattr(self, orientation + "_strategy")
if strategy not in TABLE_STRATEGIES:
raise ValueError(
f"{orientation}_strategy must be one of"
f'{{{",".join(TABLE_STRATEGIES)}}}'
)
if self.text_settings is None:
self.text_settings = {}
# This next section is for backwards compatibility
for attr in ["x_tolerance", "y_tolerance"]:
if attr not in self.text_settings:
self.text_settings[attr] = self.text_settings.get("tolerance", 3)
if "tolerance" in self.text_settings:
del self.text_settings["tolerance"]
# End of that section
for attr, fallback in [
("snap_x_tolerance", "snap_tolerance"),
("snap_y_tolerance", "snap_tolerance"),
("join_x_tolerance", "join_tolerance"),
("join_y_tolerance", "join_tolerance"),
("intersection_x_tolerance", "intersection_tolerance"),
("intersection_y_tolerance", "intersection_tolerance"),
]:
if getattr(self, attr) is UNSET:
setattr(self, attr, getattr(self, fallback))
return self
@classmethod
def resolve(cls, settings=None):
if settings is None:
return cls()
elif isinstance(settings, cls):
return settings
elif isinstance(settings, dict):
core_settings = {}
text_settings = {}
for k, v in settings.items():
if k[:5] == "text_":
text_settings[k[5:]] = v
else:
core_settings[k] = v
core_settings["text_settings"] = text_settings
return cls(**core_settings)
else:
raise ValueError(f"Cannot resolve settings: {settings}")
class TableFinder:
"""
Given a PDF page, find plausible table structures.
Largely borrowed from Anssi Nurminen's master's thesis:
http://dspace.cc.tut.fi/dpub/bitstream/handle/123456789/21520/Nurminen.pdf?sequence=3
... and inspired by Tabula:
https://github.com/tabulapdf/tabula-extractor/issues/16
"""
def __init__(self, page, settings=None):
self.page = weakref.proxy(page)
self.textpage = None
self.settings = TableSettings.resolve(settings)
self.edges = self.get_edges()
self.intersections = edges_to_intersections(
self.edges,
self.settings.intersection_x_tolerance,
self.settings.intersection_y_tolerance,
)
self.cells = intersections_to_cells(self.intersections)
self.tables = [
Table(self.page, cell_group)
for cell_group in cells_to_tables(self.page, self.cells)
]
def get_edges(self) -> list:
settings = self.settings
for orientation in ["vertical", "horizontal"]:
strategy = getattr(settings, orientation + "_strategy")
if strategy == "explicit":
lines = getattr(settings, "explicit_" + orientation + "_lines")
if len(lines) < 2:
raise ValueError(
f"If {orientation}_strategy == 'explicit', "
f"explicit_{orientation}_lines "
f"must be specified as a list/tuple of two or more "
f"floats/ints."
)
v_strat = settings.vertical_strategy
h_strat = settings.horizontal_strategy
if v_strat == "text" or h_strat == "text":
words = extract_words(CHARS, **(settings.text_settings or {}))
else:
words = []
v_explicit = []
for desc in settings.explicit_vertical_lines or []:
if isinstance(desc, dict):
for e in obj_to_edges(desc):
if e["orientation"] == "v":
v_explicit.append(e)
else:
v_explicit.append(
{
"x0": desc,
"x1": desc,
"top": self.page.rect[1],
"bottom": self.page.rect[3],
"height": self.page.rect[3] - self.page.rect[1],
"orientation": "v",
}
)
if v_strat == "lines":
v_base = filter_edges(EDGES, "v")
elif v_strat == "lines_strict":
v_base = filter_edges(EDGES, "v", edge_type="line")
elif v_strat == "text":
v_base = words_to_edges_v(words, word_threshold=settings.min_words_vertical)
elif v_strat == "explicit":
v_base = []
else:
v_base = []
v = v_base + v_explicit
h_explicit = []
for desc in settings.explicit_horizontal_lines or []:
if isinstance(desc, dict):
for e in obj_to_edges(desc):
if e["orientation"] == "h":
h_explicit.append(e)
else:
h_explicit.append(
{
"x0": self.page.rect[0],
"x1": self.page.rect[2],
"width": self.page.rect[2] - self.page.rect[0],
"top": desc,
"bottom": desc,
"orientation": "h",
}
)
if h_strat == "lines":
h_base = filter_edges(EDGES, "h")
elif h_strat == "lines_strict":
h_base = filter_edges(EDGES, "h", edge_type="line")
elif h_strat == "text":
h_base = words_to_edges_h(
words, word_threshold=settings.min_words_horizontal
)
elif h_strat == "explicit":
h_base = []
else:
h_base = []
h = h_base + h_explicit
edges = list(v) + list(h)
edges = merge_edges(
edges,
snap_x_tolerance=settings.snap_x_tolerance,
snap_y_tolerance=settings.snap_y_tolerance,
join_x_tolerance=settings.join_x_tolerance,
join_y_tolerance=settings.join_y_tolerance,
)
return filter_edges(edges, min_length=settings.edge_min_length)
def __getitem__(self, i):
tcount = len(self.tables)
if i >= tcount:
raise IndexError("table not on page")
while i < 0:
i += tcount
return self.tables[i]
"""
Start of PyMuPDF interface code.
The following functions are executed when "page.find_tables()" is called.
* make_chars: Fills the CHARS list with text character information extracted
via "rawdict" text extraction. Items in CHARS are formatted
as expected by the table code.
* make_edges: Fills the EDGES list with vector graphic information extracted
via "get_drawings". Items in EDGES are formatted as expected
by the table code.
The lists CHARS and EDGES are used to replace respective document access
of pdfplumber or, respectively pdfminer.
The table code has been modified to use these lists instead of accessing
page information themselves.
"""
# -----------------------------------------------------------------------------
# Extract all page characters to fill the CHARS list
# -----------------------------------------------------------------------------
def make_chars(page, clip=None):
"""Extract text as "rawdict" to fill CHARS."""
chars = CHARS._list() # bind once: avoid per-append proxy overhead below
page_number = page.number + 1
page_height = page.rect.height
ctm = page.transformation_matrix
TEXTPAGE = page.get_textpage(clip=clip, flags=FLAGS)
blocks = page.get_text("rawdict", textpage=TEXTPAGE)["blocks"]
doctop_base = page_height * page.number
for block in blocks:
for line in block["lines"]:
ldir = line["dir"] # = (cosine, sine) of angle
ldir = (round(ldir[0], 4), round(ldir[1], 4))
matrix = pymupdf.Matrix(ldir[0], -ldir[1], ldir[1], ldir[0], 0, 0)
if ldir[1] == 0:
upright = True
else:
upright = False
for span in sorted(line["spans"], key=lambda s: s["bbox"][0]):
fontname = span["font"]
fontsize = span["size"]
span_bold = bool(
span["flags"] & pymupdf.TEXT_FONT_BOLD or span["char_flags"] & 8
)
color = pymupdf.sRGB_to_pdf(span["color"])
for char in sorted(span["chars"], key=lambda c: c["bbox"][0]):
bbox = pymupdf.Rect(char["bbox"])
bbox_ctm = bbox * ctm
origin = pymupdf.Point(char["origin"]) * ctm
matrix.e = origin.x
matrix.f = origin.y
text = char["c"]
char_dict = {
"adv": bbox.x1 - bbox.x0 if upright else bbox.y1 - bbox.y0,
"bottom": bbox.y1,
"doctop": bbox.y0 + doctop_base,
"fontname": fontname,
"height": bbox.y1 - bbox.y0,
"matrix": tuple(matrix),
"ncs": "DeviceRGB",
"non_stroking_color": color,
"non_stroking_pattern": None,
"object_type": "char",
"page_number": page_number,
"size": fontsize if upright else bbox.y1 - bbox.y0,
"stroking_color": color,
"stroking_pattern": None,
"bold": span_bold,
"text": text,
"top": bbox.y0,
"upright": upright,
"width": bbox.x1 - bbox.x0,
"x0": bbox.x0,
"x1": bbox.x1,
"y0": bbox_ctm.y0,
"y1": bbox_ctm.y1,
}
chars.append(char_dict)
return TEXTPAGE
# ------------------------------------------------------------------------
# Extract all page vector graphics to fill the EDGES list.
# We are ignoring Bézier curves completely and are converting everything
# else to lines.
# ------------------------------------------------------------------------
def make_edges(page, clip=None, tset=None, paths=None, add_lines=None, add_boxes=None):
edges = EDGES._list() # bind once: avoid per-append proxy overhead below
snap_x = tset.snap_x_tolerance
snap_y = tset.snap_y_tolerance
min_length = tset.edge_min_length
lines_strict = (
tset.vertical_strategy == "lines_strict"
or tset.horizontal_strategy == "lines_strict"
)
page_height = page.rect.height
doctop_basis = page.number * page_height
page_number = page.number + 1
prect = page.rect
if page.rotation in (90, 270):
w, h = prect.br
prect = pymupdf.Rect(0, 0, h, w)
if clip is not None:
clip = pymupdf.Rect(clip)
else:
clip = prect
def are_neighbors(r1, r2):
"""Detect whether r1, r2 are neighbors.
Defined as:
The minimum distance between points of r1 and points of r2 is not
larger than some delta.
This check supports empty rect-likes and thus also lines.
Note:
This type of check is MUCH faster than native Rect containment checks.
"""
if ( # check if x-coordinates of r1 are within those of r2
r2.x0 - snap_x <= r1.x0 <= r2.x1 + snap_x
or r2.x0 - snap_x <= r1.x1 <= r2.x1 + snap_x
) and ( # ... same for y-coordinates
r2.y0 - snap_y <= r1.y0 <= r2.y1 + snap_y
or r2.y0 - snap_y <= r1.y1 <= r2.y1 + snap_y
):
return True
# same check with r1 / r2 exchanging their roles (this is necessary!)
if (
r1.x0 - snap_x <= r2.x0 <= r1.x1 + snap_x
or r1.x0 - snap_x <= r2.x1 <= r1.x1 + snap_x
) and (
r1.y0 - snap_y <= r2.y0 <= r1.y1 + snap_y
or r1.y0 - snap_y <= r2.y1 <= r1.y1 + snap_y
):
return True
return False
def clean_graphics(npaths=None):
"""Detect and join rectangles of "connected" vector graphics."""
if npaths is None:
allpaths = page.get_drawings()
else: # accept passed-in vector graphics
allpaths = npaths[:] # paths relevant for table detection
paths = []
for p in allpaths:
# If only looking at lines, we ignore fill-only paths,
# except simulated lines (i.e. small width or height).
if (
lines_strict
and p["type"] == "f"
and p["rect"].width > snap_x
and p["rect"].height > snap_y
):
continue
paths.append(p)
# start with all vector graphics rectangles
prects = sorted(set([p["rect"] for p in paths]), key=lambda r: (r.y1, r.x0))
new_rects = [] # the final list of joined rectangles
# ----------------------------------------------------------------
# Strategy: Join rectangles that "almost touch" each other.
# Extend first rectangle with any other that is a "neighbor".
# Then move it to the final list and continue with the rest.
# ----------------------------------------------------------------
while prects: # the algorithm will empty this list
prect0 = prects[0] # copy of first rectangle (performance reasons!)
repeat = True
while repeat: # this loop extends first rect in list
repeat = False # set to true again if some other rect touches
for i in range(len(prects) - 1, 0, -1): # run backwards
if are_neighbors(prect0, prects[i]): # close enough to rect 0?
prect0 |= prects[i].tl # extend rect 0
prect0 |= prects[i].br # extend rect 0
del prects[i] # delete this rect
repeat = True # keep checking the rest
# move rect 0 over to result list if there is some text in it
if chars_in_rect(CHARS, prect0):
# contains text, so accept it as a table bbox candidate
new_rects.append(prect0)
del prects[0] # remove from rect list
return new_rects, paths
bboxes, paths = clean_graphics(npaths=paths)
def is_parallel(p1, p2):
"""Check if line is roughly axis-parallel."""
if abs(p1.x - p2.x) <= snap_x or abs(p1.y - p2.y) <= snap_y:
return True
return False
def make_line(p, p1, p2, clip):
"""Given 2 points, make a line dictionary for table detection."""
if not is_parallel(p1, p2): # only accepting axis-parallel lines
return {}
# compute the extremal values
x0 = min(p1.x, p2.x)
x1 = max(p1.x, p2.x)
y0 = min(p1.y, p2.y)
y1 = max(p1.y, p2.y)
# check for outside clip
if x0 > clip.x1 or x1 < clip.x0 or y0 > clip.y1 or y1 < clip.y0:
return {}
if x0 < clip.x0:
x0 = clip.x0 # adjust to clip boundary
if x1 > clip.x1:
x1 = clip.x1 # adjust to clip boundary
if y0 < clip.y0:
y0 = clip.y0 # adjust to clip boundary
if y1 > clip.y1:
y1 = clip.y1 # adjust to clip boundary
width = x1 - x0 # from adjusted values
height = y1 - y0 # from adjusted values
if width == height == 0:
return {} # nothing left to deal with
line_dict = {
"x0": x0,
"y0": page_height - y0,
"x1": x1,
"y1": page_height - y1,
"width": width,
"height": height,
"pts": [(x0, y0), (x1, y1)],
"linewidth": p["width"],
"stroke": True,
"fill": False,
"evenodd": False,
"stroking_color": p["color"] if p["color"] else p["fill"],
"non_stroking_color": None,
"object_type": "line",
"page_number": page_number,
"stroking_pattern": None,
"non_stroking_pattern": None,
"top": y0,
"bottom": y1,
"doctop": y0 + doctop_basis,
}
return line_dict
for p in paths:
items = p["items"] # items in this path
# if 'closePath', add a line from last to first point
if p["closePath"] and items[0][0] == "l" and items[-1][0] == "l":
items.append(("l", items[-1][2], items[0][1]))
for i in items:
if i[0] not in ("l", "re", "qu"):
continue # ignore anything else
if i[0] == "l": # a line
p1, p2 = i[1:]
line_dict = make_line(p, p1, p2, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
elif i[0] == "re":
# A rectangle: decompose into 4 lines, but filter out
# the ones that simulate a line
rect = i[1].normalize() # normalize the rectangle
if (
rect.width <= min_length and rect.width < rect.height
): # simulates a vertical line
x = abs(rect.x1 + rect.x0) / 2 # take middle value for x
p1 = pymupdf.Point(x, rect.y0)
p2 = pymupdf.Point(x, rect.y1)
line_dict = make_line(p, p1, p2, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
continue
if (
rect.height <= min_length and rect.height < rect.width
): # simulates a horizontal line
y = abs(rect.y1 + rect.y0) / 2 # take middle value for y
p1 = pymupdf.Point(rect.x0, y)
p2 = pymupdf.Point(rect.x1, y)
line_dict = make_line(p, p1, p2, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
continue
line_dict = make_line(p, rect.tl, rect.bl, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(p, rect.bl, rect.br, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(p, rect.br, rect.tr, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(p, rect.tr, rect.tl, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
else: # must be a quad
# we convert it into (up to) 4 lines
ul, ur, ll, lr = i[1]
line_dict = make_line(p, ul, ll, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(p, ll, lr, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(p, lr, ur, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(p, ur, ul, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
path = {"color": (0, 0, 0), "fill": None, "width": 1}
for bbox in bboxes: # add the border lines for all enveloping bboxes
line_dict = make_line(path, bbox.tl, bbox.tr, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(path, bbox.bl, bbox.br, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(path, bbox.tl, bbox.bl, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(path, bbox.tr, bbox.br, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
if add_lines is not None: # add user-specified lines
assert isinstance(add_lines, (tuple, list))
else:
add_lines = []
for p1, p2 in add_lines:
p1 = pymupdf.Point(p1)
p2 = pymupdf.Point(p2)
line_dict = make_line(path, p1, p2, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
if add_boxes is not None: # add user-specified rectangles
assert isinstance(add_boxes, (tuple, list))
else:
add_boxes = []
for box in add_boxes:
r = pymupdf.Rect(box)
line_dict = make_line(path, r.tl, r.bl, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(path, r.bl, r.br, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(path, r.br, r.tr, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
line_dict = make_line(path, r.tr, r.tl, clip)
if line_dict:
edges.append(line_to_edge(line_dict))
def page_rotation_set0(page):
"""Nullify page rotation.
To correctly detect tables, page rotation must be zero.
This function performs the necessary adjustments and returns information
for reverting this changes.
"""
mediabox = page.mediabox
rot = page.rotation # contains normalized rotation value
# need to derotate the page's content
mb = page.mediabox # current mediabox
if rot == 90:
# before derotation, shift content horizontally
mat0 = pymupdf.Matrix(1, 0, 0, 1, mb.y1 - mb.x1 - mb.x0 - mb.y0, 0)
elif rot == 270:
# before derotation, shift content vertically
mat0 = pymupdf.Matrix(1, 0, 0, 1, 0, mb.x1 - mb.y1 - mb.y0 - mb.x0)
else:
mat0 = pymupdf.Matrix(1, 0, 0, 1, -2 * mb.x0, -2 * mb.y0)
# prefix with derotation matrix
mat = mat0 * page.derotation_matrix
cmd = b"%g %g %g %g %g %g cm " % tuple(mat)
xref = pymupdf.TOOLS._insert_contents(page, cmd, 0)
# swap x- and y-coordinates
if rot in (90, 270):
x0, y0, x1, y1 = mb
mb.x0 = y0
mb.y0 = x0
mb.x1 = y1
mb.y1 = x1
page.set_mediabox(mb)
page.set_rotation(0)
# refresh the page to apply these changes
doc = page.parent
pno = page.number
page = doc[pno]
return page, xref, rot, mediabox
def page_rotation_reset(page, xref, rot, mediabox):
"""Reset page rotation to original values.
To be used before we return tables."""
doc = page.parent # document of the page
doc.update_stream(xref, b" ") # remove de-rotation matrix
page.set_mediabox(mediabox) # set mediabox to old value
page.set_rotation(rot) # set rotation to old value
pno = page.number
page = doc[pno] # update page info
return page
# ---------------------------------------------------------------------------
# find_tables(refine=True) reconstruction glue.
#
# Resolve a merged-cell placement grid (falling back to a flat 1x1 grid when
# span resolution changes the column count), determine the header/body split,
# and tag each placement td/th. Built on resolve_spans, find_header_region and
# the refine_* stages; only reached when refine=True.
# ---------------------------------------------------------------------------
def _refine_placement_grid_width(placements):
"""Column extent of a placement grid after resolving colspan/rowspan."""
occupied = set()
max_col = 0
for row_idx, row in enumerate(placements):
col_idx = 0
for placement in row:
while (row_idx, col_idx) in occupied:
col_idx += 1
for dr in range(placement.rowspan):
for dc in range(placement.colspan):
occupied.add((row_idx + dr, col_idx + dc))
col_idx += placement.colspan
max_col = max(max_col, col_idx)
return max_col
def _refine_flat_placement_grid(page, cells, col_count):
"""Flat fallback grid: one 1x1 SpanCell per slot, padded to ``col_count``.
Selects words per cell by center-point and synthesizes each cell's line text
("" for a gap), using the same word source and line builder as resolve_spans.
Used when span resolution changes the column count."""
page_words = _refine_page_words(page)
grid = []
for row in cells:
out = []
for cell in row:
if cell is None:
out.append(SpanCell(bbox=None, text="", colspan=1, rowspan=1))
else:
rect = pymupdf.Rect(cell)
line_words = [
_span_word_line_tuple(word)
for _, word in _span_select_words_in_rect(page_words, rect)
]
out.append(
SpanCell(
bbox=tuple(rect),
text=_span_words_to_line_text(line_words),
colspan=1,
rowspan=1,
)
)
while len(out) < col_count:
out.append(SpanCell(bbox=None, text="", colspan=1, rowspan=1))
grid.append(out)
return grid
def _refine_placement_or_flat_grid(page, cells, *, strict_colspan=False, header_row_count=None):
"""The reconstructed cell grid: resolved SpanCell placements, or the flat 1x1
fallback when span resolution changes the column count (grid width != col
count). ``strict_colspan``/``header_row_count`` pass straight to resolve_spans."""
col_count = max((len(row) for row in cells), default=0)
placements = resolve_spans(
page, cells, header_row_count=header_row_count, strict_colspan=strict_colspan
)
if _refine_placement_grid_width(placements) == col_count:
return placements
return _refine_flat_placement_grid(page, cells, col_count)
def _refine_placements_text_grid(grid):
"""Row-major whitespace-collapsed cell text -- the ``[[text]]`` header rules read."""
return [[collapse_cell_ws(cell.text) for cell in row] for row in grid]
def _refine_tag_grid(grid, top_header_rows):
"""Set each placement's HTML tag in place: cells in the top header rows
become ``th``; every other cell becomes ``td``."""
for row_idx, row in enumerate(grid):
for cell in row:
cell.tag = "th" if row_idx < top_header_rows else "td"
return grid
def _refine_body_start_row(page, cells):
"""Header/body boundary: resolve the merge-preserved placement grid once and
ask the header finder how many leading rows are header, clamped to [1, rows]."""
try:
model_grid = _refine_placement_or_flat_grid(page, cells)
region = find_header_region(_refine_placements_text_grid(model_grid))
except Exception:
return 1
raw = region.top_header_rows
return max(1, min(int(raw), len(cells))) if cells else 0
def _refine_build_placements(page, working, body_start):
"""Resolve the final placement grid (strict colspan, header boundary known),
run header rules on its own text grid, tag cells -> (tagged grid, region)."""
grid = _refine_placement_or_flat_grid(
page, working, strict_colspan=True, header_row_count=body_start
)
region = find_header_region(_refine_placements_text_grid(grid))
tagged = _refine_tag_grid(grid, region.top_header_rows)
return tagged, region
def find_tables(
page,
clip=None,
vertical_strategy: str = "lines",
horizontal_strategy: str = "lines",
vertical_lines: list = None,
horizontal_lines: list = None,
snap_tolerance: float = DEFAULT_SNAP_TOLERANCE,
snap_x_tolerance: float = None,
snap_y_tolerance: float = None,
join_tolerance: float = DEFAULT_JOIN_TOLERANCE,
join_x_tolerance: float = None,
join_y_tolerance: float = None,
edge_min_length: float = 3,
min_words_vertical: float = DEFAULT_MIN_WORDS_VERTICAL,
min_words_horizontal: float = DEFAULT_MIN_WORDS_HORIZONTAL,
intersection_tolerance: float = 3,
intersection_x_tolerance: float = None,
intersection_y_tolerance: float = None,
text_tolerance=3,
text_x_tolerance=3,
text_y_tolerance=3,
strategy=None, # offer abbreviation
add_lines=None, # user-specified lines
add_boxes=None, # user-specified rectangles
paths=None, # accept vector graphics as parameter
use_layout: bool = True, # gate line-based tables by layout table boxes
union: bool = False, # opt-in: fuse layout grids with line-based candidates
refine: bool = False, # opt-in: refine each detected table's cell grid
):
"""Detect and extract tables on a page.
use_layout: if True (default), use page.get_layout() to find layout-
identified table regions and use them to restrict/complete line-based
detection; if layout ran and found no tables, return an empty result
immediately. Set False to always run full detection.
union: if True (requires the layout analyzer), detect tables by fusing the
layout analyzer's table grids with the line-based finder's candidates -- a
candidate matching a layout table 1:1 can replace its grid, candidates
contained in one layout table can split it, and unowned candidates are
appended -- returning the fused tables in layout order then appended order.
Off by default. Needs the raw layout form: when page.layout_information is
None it is computed with page.get_layout(return_raw=True); when already
populated it is reused as-is.
refine: if True, refine each detected table's cell grid before building the
Table -- splitting rows/columns the line grid merged (using page text and
background shading) -- then reconstruct its merged-cell structure and header
semantics. The tagged grid is attached as Table.placements (a row-major grid
of SpanCell colspan/rowspan placements, each carrying its td/th tag; None on
the default path), the header meta as Table.header_rows/section_rows, and
Table.to_html() serializes it. Off by default. A grid whose column count the
span resolution cannot preserve falls back to a flat one-cell-per-slot
placement grid.
"""
pymupdf._warn_layout_once()
_CHARS_VAR.set([])
_EDGES_VAR.set([])
old_small = bool(pymupdf.TOOLS.set_small_glyph_heights()) # save old value
pymupdf.TOOLS.set_small_glyph_heights(True) # we need minimum bboxes
if page.rotation != 0:
page, old_xref, old_rot, old_mediabox = page_rotation_set0(page)
else:
old_xref, old_rot, old_mediabox = None, None, None
if snap_x_tolerance is None:
snap_x_tolerance = UNSET
if snap_y_tolerance is None:
snap_y_tolerance = UNSET
if join_x_tolerance is None:
join_x_tolerance = UNSET
if join_y_tolerance is None:
join_y_tolerance = UNSET
if intersection_x_tolerance is None:
intersection_x_tolerance = UNSET
if intersection_y_tolerance is None:
intersection_y_tolerance = UNSET
if strategy is not None:
vertical_strategy = strategy
horizontal_strategy = strategy
settings = {
"vertical_strategy": vertical_strategy,
"horizontal_strategy": horizontal_strategy,
"explicit_vertical_lines": vertical_lines,
"explicit_horizontal_lines": horizontal_lines,
"snap_tolerance": snap_tolerance,
"snap_x_tolerance": snap_x_tolerance,
"snap_y_tolerance": snap_y_tolerance,
"join_tolerance": join_tolerance,
"join_x_tolerance": join_x_tolerance,
"join_y_tolerance": join_y_tolerance,
"edge_min_length": edge_min_length,
"min_words_vertical": min_words_vertical,
"min_words_horizontal": min_words_horizontal,
"intersection_tolerance": intersection_tolerance,
"intersection_x_tolerance": intersection_x_tolerance,
"intersection_y_tolerance": intersection_y_tolerance,
"text_tolerance": text_tolerance,
"text_x_tolerance": text_x_tolerance,
"text_y_tolerance": text_y_tolerance,
}
old_quad_corrections = pymupdf.TOOLS.unset_quad_corrections()
try:
if union:
# Union path: fuse layout grids with line-based candidates instead
# of the ordinary use_layout gating. It builds its own tables (and,
# via a nested find_tables, its own char list), so the make_chars/
# make_edges/TableFinder setup below is skipped; TEXTPAGE comes from
# the nested finder. Imported here, not at module top, to avoid an
# import cycle: _table_union imports find_tables back from this module.
from pymupdf._table_union import _find_tables_union
tbf = _find_tables_union(page)
TEXTPAGE = tbf.textpage
else:
boxes = []
if use_layout:
page.get_layout()
if page.layout_information:
pymupdf.TOOLS.unset_quad_corrections(True)
boxes = [
pymupdf.Rect(b[:4]) for b in page.layout_information if b[-1] == "table"
]
if not boxes and page.layout_information is not None:
# layout was executed but found no tables
# make sure we exit quickly with an empty TableFinder
tbf = TableFinder(page)
return tbf
tset = TableSettings.resolve(settings=settings)
page.table_settings = tset
TEXTPAGE = make_chars(page, clip=clip) # create character list of page
make_edges(
page,
clip=clip,
tset=tset,
paths=paths,
add_lines=add_lines,
add_boxes=add_boxes,
) # create lines and curves
tbf = TableFinder(page, settings=tset)
tbf.textpage = TEXTPAGE # store textpage for later use
if boxes:
# only keep Finder tables that match a layout box
tbf.tables = [
tab
for tab in tbf.tables
if any(_iou(tab.bbox, r) >= 0.6 for r in boxes)
]
# build the complementary list of layout table boxes
my_boxes = [
r for r in boxes if all(_iou(r, tab.bbox) < 0.6 for tab in tbf.tables)
]
if my_boxes:
word_rects = [pymupdf.Rect(w[:4]) for w in TEXTPAGE.extractWORDS()]
tp2 = page.get_textpage(flags=TABLE_DETECTOR_FLAGS)
for rect in my_boxes:
cells = make_table_from_bbox(tp2, word_rects, rect) # pylint: disable=E0606
if cells:
tbf.tables.append(Table(page, cells))
if refine:
# Grid refinement + reconstruction. Runs while the page is still
# derotated (before the finally block resets rotation) so word and
# vector coordinates match the detected cells. Order: structural
# split (shaded rows + under-segmented columns), resolve the header/
# body boundary, split over-merged body rows, resolve the final
# merged-cell placement grid (strict colspan) and tag each placement
# td/th. The tagged grid is attached as .placements, the header meta
# as .header_rows/.section_rows.
refined_tables = []
for tab in tbf.tables:
grid = _refine_cells_to_grid(tab.cells)
# The reported bbox (a union grid-ref table's layout box, else
# the cells' union) bounds the shaded-rectangle search.
working = refine_grid_structure(page, grid, table_bbox=tab.bbox)
body_start = _refine_body_start_row(page, working)
working = refine_grid_rows(page, working, header_row_count=body_start)
flat = _refine_grid_to_cells(working)
# Preserve an explicit reported-bbox override (union grid-ref
# tables): the refined grid must not change the reported region.
new_tab = Table(page, flat, bbox=tab._bbox) if flat else tab
# Build the tagged model on `working` directly, not on the
# re-gridded new_tab.cells, so placements match the refined grid.
placements, region = _refine_build_placements(page, working, body_start)
new_tab.placements = placements
new_tab.header_rows = region.top_header_rows
new_tab.section_rows = region.section_header_rows
refined_tables.append(new_tab)
tbf.tables = refined_tables
except Exception as e:
pymupdf.message("find_tables: exception occurred: %s" % str(e))
return None
finally:
pymupdf.TOOLS.set_small_glyph_heights(old_small)
if old_xref is not None:
page = page_rotation_reset(page, old_xref, old_rot, old_mediabox)
pymupdf.TOOLS.unset_quad_corrections(old_quad_corrections)
# Snapshot the page's characters onto each table so a later find_tables()
# call's CHARS reset cannot leak into this table's extract(). One shallow
# copy shared per call.
chars = list(CHARS)
for table in tbf.tables:
table.textpage = TEXTPAGE
table._chars = chars
return tbf
|