Tso tawm cov ntawv nyeem qhov system Tesseract 5.0

Kev tso tawm ntawm Tesseract 4.1 optical text recognition system tau luam tawm, txhawb kev lees paub ntawm UTF-8 cov cim thiab cov ntawv hauv ntau dua 100 hom lus, suav nrog Lavxias, Kazakh, Belarusian thiab Ukrainian. Cov txiaj ntsig tuaj yeem khaws cia hauv cov ntawv nyeem dawb lossis hauv HTML (hOCR), ALTO (XML), PDF thiab TSV hom. Lub kaw lus tau tsim nyob rau hauv 1985-1995 hauv Hewlett Packard laboratory; hauv 2005, cov cai tau qhib raws li daim ntawv tso cai Apache thiab tau tsim ntxiv nrog kev koom tes ntawm Google cov neeg ua haujlwm. Qhov chaws ntawm qhov project yog muab faib raws li Apache 2.0 daim ntawv tso cai.

Tesseract suav nrog kev siv hluav taws xob console thiab lub tsev qiv ntawv libtesseract rau embedding OCR functionality rau lwm yam kev siv. Thib peb GUI interfaces uas txhawb Tesseract suav nrog gImageReader, VietOCR thiab YAGF. Muaj ob lub cav paub txog: ib qho classic uas lees paub cov ntawv nyob rau theem ntawm tus kheej tus cwj pwm qauv, thiab ib qho tshiab raws li kev siv tshuab kev kawm raws li LSTM recurrent neural network, optimized rau paub tag nrho cov hlua thiab tso cai rau ib tug nce qhov tseeb ntawm qhov tseeb. Cov qauv npaj ua tiav tau luam tawm rau 123 yam lus. Txhawm rau txhim kho kev ua tau zoo, cov qauv siv OpenMP thiab SIMD cov lus qhia AVX2, AVX, NEON lossis SSE4.1 muaj.

Kev txhim kho loj hauv Tesseract 5.0:

  • Ib qho kev hloov pauv tseem ceeb hauv tus lej version yog vim muaj kev hloov pauv rau API uas ua txhaum kev sib raug zoo. Tshwj xeeb tshaj yog, cov ntaub ntawv muaj nyob rau hauv libtesseract API tsis raug khi rau tus tswv GenericVector thiab STRING cov ntaub ntawv hom, hauv kev pom zoo ntawm std::string thiab std::vector.
  • Lub hauv paus ntawv ntoo tau reorganized. Cov ntaub ntawv header pej xeem tau raug tsiv mus rau hauv cov npe suav nrog / cov ntawv teev npe.
  • Kev tswj hwm nco tau raug kho dua tshiab, tag nrho cov malloc thiab hu dawb tau hloov nrog C ++ code. Ib qho kev hloov kho tshiab ntawm txoj cai tau ua tiav.
  • Ntxiv kev ua kom zoo rau ARM thiab ARM64 architectures; ARM NEON cov lus qhia yog siv los ua kom cov lej suav. Kev ua tau zoo ntawm kev ua haujlwm zoo ib yam rau txhua qhov architectures tau ua tiav.
  • Cov qauv tshiab rau cov qauv kev cob qhia thiab kev paub cov ntawv nyeem raws li kev siv cov ntsiab lus ntab ntab tau ua tiav. Cov qauv tshiab muaj kev ua tau zoo dua thiab kev siv lub cim xeeb qis dua. Hauv LSTM lub cav, float32 ceev hom yog qhib los ntawm lub neej ntawd.
  • Kev hloov pauv tau ua rau siv Unicode normalization siv NFC (Normalization Form Canonical) daim ntawv.
  • Ntxiv ib qho kev xaiv los teeb tsa lub cav nthuav dav (--loglevel).
  • Lub kaw lus tsim raws li Autotools tau raug kho dua tshiab thiab hloov mus tsim hauv hom tsis rov ua dua.
  • Cov ceg "tus tswv" hauv Git tau hloov npe mus rau "lub ntsiab".
  • Ntxiv kev txhawb nqa rau kev tshaj tawm tshiab ntawm macOS thiab Apple systems raws li M1 nti.

    Tau qhov twg los: opennet.ru

Ntxiv ib saib