Ho lokolloa ha Tesseract 4.1 optical text recognition system e hatisitsoe, e tšehetsang tlhokomelo ea litlhaku tsa UTF-8 le litemana ka lipuo tse fetang 100, ho akarelletsa Serussia, Kazakh, Belarusian le Seukraine. Sephetho se ka bolokoa ka mongolo o hlakileng kapa ka liforomo tsa HTML (hOCR), ALTO (XML), PDF le TSV. Sistimi e qalile ka 1985-1995 ka laboratoring ea Hewlett Packard; ka 2005, khoutu e ile ea buloa tlasa laesense ea Apache mme ea ntlafatsoa hape ka ho nka karolo ha basebetsi ba Google. Khoutu ea mohloli oa projeke e ajoa tlasa laesense ea Apache 2.0.
Tesseract e kenyelletsa sesebelisoa sa console le laeborari ea libtesseract bakeng sa ho kenya ts'ebetso ea OCR lits'ebetsong tse ling. Likhokahano tsa mokha oa boraro tsa GUI tse tšehetsang Tesseract li kenyelletsa gImageReader, VietOCR le YAGF. Ho fanoa ka lienjineri tse peli tsa temohisiso: ea khale e amohelang mongolo boemong ba lipaterone tsa motho ka mong, le e ncha e thehiloeng ho ts'ebeliso ea sistimi ea ho ithuta ka mochini e thehiloeng ho netweke ea LSTM e pheta-phetoang ea neural, e ntlafalitsoeng bakeng sa ho lemoha likhoele tse felletseng le ho lumella ho keketseho e kholo ea ho nepahala. Mefuta e seng e lokisitsoe e seng e hatisitsoe ka lipuo tse 123. Ho ntlafatsa ts'ebetso, ho fanoa ka li-module tse sebelisang OpenMP le SIMD AVX2, AVX, NEON kapa SSE4.1.
Lintlafatso tse kholo ho Tesseract 5.0:
- Phetoho e kholo ea nomoro ea phetolelo e bakoa ke liphetoho tse entsoeng ho API tse senyang tšebelisano. Haholo-holo, libtesseract API e fumanehang phatlalatsa ha e sa tlamahanngoa le mefuta ea data ea GenericVector le STRING, molemong oa std::string le std::vector.
- Sefate sa mongolo oa mohloli se hlophisitsoe bocha. Lifaele tsa lihlooho tsa sechaba li fetiselitsoe ho kenyeletsa/tesseract directory.
- Tsamaiso ea memori e hlophisitsoe bocha, mehala eohle ea malloc le ea mahala e nkeloe sebaka ke khoutu ea C++. Ho se ho entsoe ntlafatso e akaretsang ea khoutu.
- Lintlafatso tse kenyellelitsoeng bakeng sa meralo ea ARM le ARM64; Litaelo tsa ARM NEON li sebelisoa ho potlakisa lipalo. Ntlafatso ea ts'ebetso e tloaelehileng ho meralo eohle ea meralo e se e entsoe.
- Ho kentsoe tšebetsong mekhoa e mecha ea lithupelo le temoho ea mongolo e ipapisitseng le ts'ebeliso ea lipalo tsa lintlha tse phaphametseng. Mefuta e mecha e fana ka ts'ebetso e phahameng le tšebeliso e tlase ea memori. Ka enjene ea LSTM, mokhoa o potlakileng oa float32 o lumelloa ka mokhoa oa kamehla.
- Ho entsoe phetoho ea ho sebelisa Unicode normalization ho sebelisoa foromo ea NFC (Normalization Form Canonical).
- E kentse khetho ea ho hlophisa lintlha tsa log (--loglevel).
- Sistimi ea kaho e thehiloeng ho Autotools e hlophisitsoe bocha mme e fetotsoe hore e hahe ka mokhoa o sa fetoheng.
- Lekala la "master" ho Git le rehiloe lebitso la "main".
- Ts'ehetso e ekelitsoeng bakeng sa lintlafatso tse ncha tsa litsamaiso tsa macOS le Apple tse thehiloeng ho chip ea M1.
Source: opennet.ru