1TESSERACT(1) 2============ 3:doctype: manpage 4 5NAME 6---- 7tesseract - command-line OCR engine 8 9SYNOPSIS 10-------- 11*tesseract* 'FILE' 'OUTPUTBASE' ['OPTIONS']... ['CONFIGFILE']... 12 13DESCRIPTION 14----------- 15tesseract(1) is a commercial quality OCR engine originally developed at HP 16between 1985 and 1995. In 1995, this engine was among the top 3 evaluated by 17UNLV. It was open-sourced by HP and UNLV in 2005, and has been developed 18at Google since then. 19 20 21IN/OUT ARGUMENTS 22---------------- 23'FILE':: 24 The name of the input file. 25 This can either be an image file or a text file. + 26 Most image file formats (anything readable by Leptonica) are supported. + 27 A text file lists the names of all input images (one image name per line). 28 The results will be combined in a single file for each output file format 29 (txt, pdf, hocr, xml). + 30 If 'FILE' is `stdin` or `-` then the standard input is used. 31 32'OUTPUTBASE':: 33 The basename of the output file (to which the appropriate extension 34 will be appended). By default the output will be a text file 35 with `.txt` added to the basename unless there are one or more 36 parameters set which explicitly specify the desired output. + 37 If 'OUTPUTBASE' is `stdout` or `-` then the standard output is used. 38 39 40[[TESSDATADIR]] 41OPTIONS 42------- 43*-c* 'CONFIGVAR=VALUE':: 44 Set value for parameter 'CONFIGVAR' to VALUE. Multiple *-c* arguments are allowed. 45 46*--dpi* 'N':: 47 Specify the resolution 'N' in DPI for the input image(s). 48 A typical value for 'N' is `300`. Without this option, 49 the resolution is read from the metadata included in the image. 50 If an image does not include that information, Tesseract tries to guess it. 51 52*-l* 'LANG':: 53*-l* 'SCRIPT':: 54 The language or script to use. 55 If none is specified, `eng` (English) is assumed. 56 Multiple languages may be specified, separated by plus characters. 57 Tesseract uses 3-character ISO 639-2 language codes 58 (see <<LANGUAGES,*LANGUAGES AND SCRIPTS*>>). 59 60*--psm* 'N':: 61 Set Tesseract to only run a subset of layout analysis and assume 62 a certain form of image. The options for 'N' are: 63 64 0 = Orientation and script detection (OSD) only. 65 1 = Automatic page segmentation with OSD. 66 2 = Automatic page segmentation, but no OSD, or OCR. (not implemented) 67 3 = Fully automatic page segmentation, but no OSD. (Default) 68 4 = Assume a single column of text of variable sizes. 69 5 = Assume a single uniform block of vertically aligned text. 70 6 = Assume a single uniform block of text. 71 7 = Treat the image as a single text line. 72 8 = Treat the image as a single word. 73 9 = Treat the image as a single word in a circle. 74 10 = Treat the image as a single character. 75 11 = Sparse text. Find as much text as possible in no particular order. 76 12 = Sparse text with OSD. 77 13 = Raw line. Treat the image as a single text line, 78 bypassing hacks that are Tesseract-specific. 79 80*--oem* 'N':: 81 Specify OCR Engine mode. The options for 'N' are: 82 83 0 = Original Tesseract only. 84 1 = Neural nets LSTM only. 85 2 = Tesseract + LSTM. 86 3 = Default, based on what is available. 87 88*--tessdata-dir* 'PATH':: 89 Specify the location of tessdata path. 90 91*--user-patterns* 'FILE':: 92 Specify the location of user patterns file. 93 94*--user-words* 'FILE':: 95 Specify the location of user words file. 96 97[[CONFIGFILE]] 98'CONFIGFILE':: 99 The name of a config to use. The name can be a file in `tessdata/configs` 100 or `tessdata/tessconfigs`, or an absolute or relative file path. 101 A config is a plain text file which contains a list of parameters and 102 their values, one per line, with a space separating parameter from value. + 103 Interesting config files include: 104 105 * *alto* -- Output in ALTO format ('OUTPUTBASE'`.xml`). 106 * *hocr* -- Output in hOCR format ('OUTPUTBASE'`.hocr`). 107 * *pdf* -- Output PDF ('OUTPUTBASE'`.pdf`). 108 * *tsv* -- Output TSV ('OUTPUTBASE'`.tsv`). 109 * *txt* -- Output plain text ('OUTPUTBASE'`.txt`). 110 * *get.images* -- Write processed input images to file ('OUTPUTBASE'`.processedPAGENUMBER.tif`). 111 * *logfile* -- Redirect debug messages to file (`tesseract.log`). 112 * *lstm.train* -- Output files used by LSTM training ('OUTPUTBASE'`.lstmf`). 113 * *makebox* -- Write box file ('OUTPUTBASE'`.box`). 114 * *quiet* -- Redirect debug messages to '/dev/null'. 115 116It is possible to select several config files, for example 117`tesseract image.png demo alto hocr pdf txt` will create four output files 118`demo.alto`, `demo.hocr`, `demo.pdf` and `demo.txt` with the OCR results. 119 120*Nota bene:* The options *-l* 'LANG', *-l* 'SCRIPT' and *--psm* 'N' 121must occur before any 'CONFIGFILE'. 122 123 124SINGLE OPTIONS 125-------------- 126*-h, --help*:: 127 Show help message. 128 129*--help-extra*:: 130 Show extra help for advanced users. 131 132*--help-psm*:: 133 Show page segmentation modes. 134 135*--help-oem*:: 136 Show OCR Engine modes. 137 138*-v, --version*:: 139 Returns the current version of the tesseract(1) executable. 140 141*--list-langs*:: 142 List available languages for tesseract engine. 143 Can be used with *--tessdata-dir* 'PATH'. 144 145*--print-parameters*:: 146 Print tesseract parameters. 147 148 149[[LANGUAGES]] 150LANGUAGES AND SCRIPTS 151--------------------- 152 153To recognize some text with Tesseract, it is normally necessary to specify 154the language(s) or script(s) of the text (unless it is English text which is 155supported by default) using *-l* 'LANG' or *-l* 'SCRIPT'. 156 157Selecting a language automatically also selects the language specific 158character set and dictionary (word list). 159 160Selecting a script typically selects all characters of that script 161which can be from different languages. The dictionary which is included 162also contains a mix from different languages. 163In most cases, a script also supports English. 164So it is possible to recognize a language that has not been specifically 165trained for by using traineddata for the script it is written in. 166 167More than one language or script may be specified by using `+`. 168Example: `tesseract myimage.png myimage -l eng+deu+fra`. 169 170https://github.com/tesseract-ocr/tessdata_fast provides fast language and 171script models which are also part of Linux distributions. 172 173For Tesseract 4, `tessdata_fast` includes traineddata files for the 174following languages: 175 176*afr* (Afrikaans), 177*amh* (Amharic), 178*ara* (Arabic), 179*asm* (Assamese), 180*aze* (Azerbaijani), 181*aze_cyrl* (Azerbaijani - Cyrilic), 182*bel* (Belarusian), 183*ben* (Bengali), 184*bod* (Tibetan), 185*bos* (Bosnian), 186*bre* (Breton), 187*bul* (Bulgarian), 188*cat* (Catalan; Valencian), 189*ceb* (Cebuano), 190*ces* (Czech), 191*chi_sim* (Chinese simplified), 192*chi_tra* (Chinese traditional), 193*chr* (Cherokee), 194*cos* (Corsican), 195*cym* (Welsh), 196*dan* (Danish), 197*deu* (German), 198*div* (Dhivehi), 199*dzo* (Dzongkha), 200*ell* (Greek, Modern, 1453-), 201*eng* (English), 202*enm* (English, Middle, 1100-1500), 203*epo* (Esperanto), 204*equ* (Math / equation detection module), 205*est* (Estonian), 206*eus* (Basque), 207*fas* (Persian), 208*fao* (Faroese), 209*fil* (Filipino), 210*fin* (Finnish), 211*fra* (French), 212*frk* (Frankish), 213*frm* (French, Middle, ca.1400-1600), 214*fry* (West Frisian), 215*gla* (Scottish Gaelic), 216*gle* (Irish), 217*glg* (Galician), 218*grc* (Greek, Ancient, to 1453), 219*guj* (Gujarati), 220*hat* (Haitian; Haitian Creole), 221*heb* (Hebrew), 222*hin* (Hindi), 223*hrv* (Croatian), 224*hun* (Hungarian), 225*hye* (Armenian), 226*iku* (Inuktitut), 227*ind* (Indonesian), 228*isl* (Icelandic), 229*ita* (Italian), 230*ita_old* (Italian - Old), 231*jav* (Javanese), 232*jpn* (Japanese), 233*kan* (Kannada), 234*kat* (Georgian), 235*kat_old* (Georgian - Old), 236*kaz* (Kazakh), 237*khm* (Central Khmer), 238*kir* (Kirghiz; Kyrgyz), 239*kmr* (Kurdish Kurmanji), 240*kor* (Korean), 241*kor_vert* (Korean vertical), 242*lao* (Lao), 243*lat* (Latin), 244*lav* (Latvian), 245*lit* (Lithuanian), 246*ltz* (Luxembourgish), 247*mal* (Malayalam), 248*mar* (Marathi), 249*mkd* (Macedonian), 250*mlt* (Maltese), 251*mon* (Mongolian), 252*mri* (Maori), 253*msa* (Malay), 254*mya* (Burmese), 255*nep* (Nepali), 256*nld* (Dutch; Flemish), 257*nor* (Norwegian), 258*oci* (Occitan post 1500), 259*ori* (Oriya), 260*osd* (Orientation and script detection module), 261*pan* (Panjabi; Punjabi), 262*pol* (Polish), 263*por* (Portuguese), 264*pus* (Pushto; Pashto), 265*que* (Quechua), 266*ron* (Romanian; Moldavian; Moldovan), 267*rus* (Russian), 268*san* (Sanskrit), 269*sin* (Sinhala; Sinhalese), 270*slk* (Slovak), 271*slv* (Slovenian), 272*snd* (Sindhi), 273*spa* (Spanish; Castilian), 274*spa_old* (Spanish; Castilian - Old), 275*sqi* (Albanian), 276*srp* (Serbian), 277*srp_latn* (Serbian - Latin), 278*sun* (Sundanese), 279*swa* (Swahili), 280*swe* (Swedish), 281*syr* (Syriac), 282*tam* (Tamil), 283*tat* (Tatar), 284*tel* (Telugu), 285*tgk* (Tajik), 286*tha* (Thai), 287*tir* (Tigrinya), 288*ton* (Tonga), 289*tur* (Turkish), 290*uig* (Uighur; Uyghur), 291*ukr* (Ukrainian), 292*urd* (Urdu), 293*uzb* (Uzbek), 294*uzb_cyrl* (Uzbek - Cyrilic), 295*vie* (Vietnamese), 296*yid* (Yiddish), 297*yor* (Yoruba) 298 299To use a non-standard language pack named `foo.traineddata`, set the 300`TESSDATA_PREFIX` environment variable so the file can be found at 301`TESSDATA_PREFIX/tessdata/foo.traineddata` and give Tesseract the 302argument *-l* `foo`. 303 304For Tesseract 4, `tessdata_fast` includes traineddata files for the 305following scripts: 306 307*Arabic*, 308*Armenian*, 309*Bengali*, 310*Canadian_Aboriginal*, 311*Cherokee*, 312*Cyrillic*, 313*Devanagari*, 314*Ethiopic*, 315*Fraktur*, 316*Georgian*, 317*Greek*, 318*Gujarati*, 319*Gurmukhi*, 320*HanS* (Han simplified), 321*HanS_vert* (Han simplified, vertical), 322*HanT* (Han traditional), 323*HanT_vert* (Han traditional, vertical), 324*Hangul*, 325*Hangul_vert* (Hangul vertical), 326*Hebrew*, 327*Japanese*, 328*Japanese_vert* (Japanese vertical), 329*Kannada*, 330*Khmer*, 331*Lao*, 332*Latin*, 333*Malayalam*, 334*Myanmar*, 335*Oriya* (Odia), 336*Sinhala*, 337*Syriac*, 338*Tamil*, 339*Telugu*, 340*Thaana*, 341*Thai*, 342*Tibetan*, 343*Vietnamese*. 344 345The same languages and scripts are available from 346https://github.com/tesseract-ocr/tessdata_best. 347`tessdata_best` provides slow language and script models. 348These models are needed for training. They also can give better OCR results, 349but the recognition takes much more time. 350 351Both `tessdata_fast` and `tessdata_best` only support the LSTM OCR engine. 352 353There is a third repository, https://github.com/tesseract-ocr/tessdata, 354with models which support both the Tesseract 3 legacy OCR engine and the 355Tesseract 4 LSTM OCR engine. 356 357 358CONFIG FILES AND AUGMENTING WITH USER DATA 359------------------------------------------ 360 361Tesseract config files consist of lines with parameter-value pairs (space 362separated). The parameters are documented as flags in the source code like 363the following one in tesseractclass.h: 364 365`STRING_VAR_H(tessedit_char_blacklist, "", 366 "Blacklist of chars not to recognize");` 367 368These parameters may enable or disable various features of the engine, and 369may cause it to load (or not load) various data. For instance, let's suppose 370you want to OCR in English, but suppress the normal dictionary and load an 371alternative word list and an alternative list of patterns -- these two files 372are the most commonly used extra data files. 373 374If your language pack is in '/path/to/eng.traineddata' and the hocr config 375is in '/path/to/configs/hocr' then create three new files: 376 377'/path/to/eng.user-words': 378[verse] 379the 380quick 381brown 382fox 383jumped 384 385'/path/to/eng.user-patterns': 386[verse] 3871-\d\d\d-GOOG-411 388www.\n\\\*.com 389 390'/path/to/configs/bazaar': 391[verse] 392load_system_dawg F 393load_freq_dawg F 394user_words_suffix user-words 395user_patterns_suffix user-patterns 396 397Now, if you pass the word 'bazaar' as a <<CONFIGFILE,'CONFIGFILE'>> to 398Tesseract, Tesseract will not bother loading the system dictionary nor 399the dictionary of frequent words and will load and use the 'eng.user-words' 400and 'eng.user-patterns' files you provided. The former is a simple word list, 401one per line. The format of the latter is documented in 'dict/trie.h' 402on 'read_pattern_list()'. 403 404 405ENVIRONMENT VARIABLES 406--------------------- 407*`TESSDATA_PREFIX`*:: 408 If the `TESSDATA_PREFIX` is set to a path, then that path is used to 409 find the `tessdata` directory with language and script recognition 410 models and config files. 411 Using <<TESSDATADIR,*--tessdata-dir* 'PATH'>> is the recommended alternative. 412*`OMP_THREAD_LIMIT`*:: 413 If the `tesseract` executable was built with multithreading support, 414 it will normally use four CPU cores for the OCR process. While this 415 can be faster for a single image, it gives bad performance if the host 416 computer provides less than four CPU cores or if OCR is made for many images. 417 Only a single CPU core is used with `OMP_THREAD_LIMIT=1`. 418 419 420HISTORY 421------- 422The engine was developed at Hewlett Packard Laboratories Bristol and at 423Hewlett Packard Co, Greeley Colorado between 1985 and 1994, with some more 424changes made in 1996 to port to Windows, and some $$C++$$izing in 1998. A 425lot of the code was written in C, and then some more was written in $$C++$$. 426The $$C++$$ code makes heavy use of a list system using macros. This predates 427STL, was portable before STL, and is more efficient than STL lists, but has 428the big negative that if you do get a segmentation violation, it is hard to 429debug. 430 431Version 2.00 brought Unicode (UTF-8) support, six languages, and the ability 432to train Tesseract. 433 434Tesseract was included in UNLV's Fourth Annual Test of OCR Accuracy. 435See <https://github.com/tesseract-ocr/docs/blob/main/AT-1995.pdf>. 436Since Tesseract 2.00, 437scripts are now included to allow anyone to reproduce some of these tests. 438See <https://tesseract-ocr.github.io/tessdoc/TestingTesseract.html> for more 439details. 440 441Tesseract 3.00 added a number of new languages, including Chinese, Japanese, 442and Korean. It also introduced a new, single-file based system of managing 443language data. 444 445Tesseract 3.02 added BiDirectional text support, the ability to recognize 446multiple languages in a single image, and improved layout analysis. 447 448Tesseract 4 adds a new neural net (LSTM) based OCR engine which is focused 449on line recognition, but also still supports the legacy Tesseract OCR engine of 450Tesseract 3 which works by recognizing character patterns. Compatibility with 451Tesseract 3 is enabled by `--oem 0`. This also needs traineddata files which 452support the legacy engine, for example those from the tessdata repository 453(https://github.com/tesseract-ocr/tessdata). 454 455For further details, see the release notes in the Tesseract documentation 456(<https://tesseract-ocr.github.io/tessdoc/ReleaseNotes.html>). 457 458 459RESOURCES 460--------- 461Main web site: <https://github.com/tesseract-ocr> + 462User forum: <https://groups.google.com/g/tesseract-ocr> + 463Documentation: <https://tesseract-ocr.github.io/> + 464Information on training: <https://tesseract-ocr.github.io/tessdoc/Training-Tesseract.html> 465 466SEE ALSO 467-------- 468ambiguous_words(1), cntraining(1), combine_tessdata(1), dawg2wordlist(1), 469shape_training(1), mftraining(1), unicharambigs(5), unicharset(5), 470unicharset_extractor(1), wordlist2dawg(1) 471 472AUTHOR 473------ 474Tesseract development was led at Hewlett-Packard and Google by Ray Smith. 475The development team has included: 476 477Ahmad Abdulkader, Chris Newton, Dan Johnson, Dar-Shyang Lee, David Eger, 478Eric Wiseblatt, Faisal Shafait, Hiroshi Takenaka, Joe Liu, Joern Wanke, 479Mark Seaman, Mickey Namiki, Nicholas Beato, Oded Fuhrmann, Phil Cheatle, 480Pingping Xiu, Pong Eksombatchai (Chantat), Ranjith Unnikrishnan, Raquel 481Romano, Ray Smith, Rika Antonova, Robert Moss, Samuel Charron, Sheelagh 482Lloyd, Shobhit Saxena, and Thomas Kielbus. 483 484For a list of contributors see 485<https://github.com/tesseract-ocr/tesseract/blob/main/AUTHORS>. 486 487COPYING 488------- 489Licensed under the Apache License, Version 2.0 490
