Tshiab kho qhov muag ntawv nyeem system EasyOCR

Qhov project EasyOCR Ib qho tshiab optical text recognition system yog tsim los uas txhawb ntau tshaj 40 hom lus, suav nrog Askiv, German, Fabkis, Nyiv, Suav, Kauslim, Uzbek, Azerbaijani thiab Lithuanian. Cyrillic-raws li hom lus tseem tsis tau txais kev txhawb nqa, tab sis lawv tau muab ntxiv rau hauv cov npe ntawm cov phiaj xwm. Cov cai sau hauv Python siv lub moj khaum PyTorch и faib los ntawm muaj ntawv tso cai nyob rau hauv Apache 2.0. Rau kev thauj khoom muab npaj ua qauv rau cov lus raws li cov tsiaj ntawv Latin thiab hieroglyphs.

Kev kawm tshuab yog siv los txheeb xyuas thiab paub cov ntawv hauv ib daim duab. Lub tshuab kawm algorithm yog siv los txheeb xyuas cov ntawv Khoom siv tes ua (Character-Region Awareness For Text) hauv kev siv rau PyTorch, muaj peev xwm ua kom pom cov ntawv ntawm cov khoom siv tsis raug, suav nrog cov ntawv sau, cov ntaub ntawv qhia thiab cov paib txoj kev. Lub convolutional recurrent neural network yog siv los paub txog cov cim ua ntu zus CRNN (Convolutional Recurrent Neural Network, ua ke ntawm DCNN thiab RNN) thiab algorithm CTC BeamSearch CTC BeamSearch (Connectionist Temporal Classification) los txiav txim siab neural network tso tawm rau hauv cov ntawv sawv cev.

Tau qhov twg los: opennet.ru