I-VisionLabs, inxalenye ye-ecosystem ye-Sberbank, yaphuma phezulu okwesibini ekuvavanyeni i-algorithms yokuqaphela ubuso kwi-US National Institute of Standards and Technology (NIST).
Itekhnoloji yeVisionLabs iphumelele indawo yokuqala kudidi lweMugshot kwaye yangena kwi-3 ephezulu kudidi lweVisa. Ngokwesantya sokuqaphela, i-algorithm yayo iphindwe kabini ngokukhawuleza njengezisombululo ezifanayo zabanye abathathi-nxaxheba. Ngexesha lokhuphiswano, ngaphezulu kwe-algorithms ye-100 evela kubaboneleli abahlukeneyo baye bavavanywa.
I-NIST iqalise uvavanyo olutsha lweetekhnoloji zokuqaphela ubuso ngoFebruwari 2017. Uvavanyo lwe-FRVT 1: 1 luhambelana nemeko yokuqinisekisa isazisi somntu ngokuqinisekiswa kwefoto. Uphononongo, ngakumbi, lunceda iSebe lezoRhwebo lase-US ukuchonga abona baboneleli bezisombululo behlabathi kweli candelo lesoftware.
Kudidi lweMugshot (ifoto yesaphuli-mthetho, apho ukukhanya kunye nemvelaphi iyaguquguquka kwaye umgangatho womfanekiso unokuba mbi), ukuqondwa kobuso kuvavanywa kwisiseko sedatha yeefoto ezingaphezu kwesigidi esinye sabantu. Iqulethe iifoto zomntu omnye onomahluko omkhulu weminyaka, okwandisa ubunzima bomsebenzi.
I-algorithm ye-VisionLabs ibona ngokuchanekileyo i-99,6% kunye nezinga elihle lobuxoki le-0,001%, eliphakamileyo kuneziphumo zabanye abathathi-nxaxheba. Uvavanyo olwahlukileyo lwacetywa kolu didi, olunikezela ukuqaphela abantu kwiifoto ezithathwe ngokwahlukeneyo kwiminyaka eli-14. Kolu vavanyo, i-VisionLabs ithathe indawo yokuqala (i-99,5% enezinga elingeyonyani le-0,001%), lizahlula njenge-algorithm yokuqaphela ubuso obunganyangekiyo.
Kudidi lweVisa (iifoto zestudio ekukhanyeni okulungileyo kwimvelaphi emhlophe), ukuqondwa kwenzeka ngokusekwe kwisiseko sedatha samakhulu amawaka eefoto zabantu. Ubunzima apha yayikukuba idatabase iqulethe iifoto zabantu abavela kumazwe angaphezu kwe-100. Kule meko, i-algorithm ye-VisionLabs ibona ngokuchanekileyo i-99,5% kunye nesantya sobuxoki se-0,0001%, sibeka indawo yesibini phakathi kwabo bonke abathengisi.
NgoAprili ka-2019, iVisionLabs yathatha indawo yokuqala kwiindidi zeMugshot, kwaye yayiphakathi kwabathathu abaphezulu kudidi lweVisa.
Ngo-Matshi ka-2019, i-VisionLabs yathatha indawo yokuqala kukhuphiswano olukhulu lwamazwe ngamazwe i-ChaLearn Face Anti-spoofing Attack Detection Challenge evela kwinkomfa ye-CVPR ka-2019, umsitho wonyaka ophambili kumbono wekhompyutha.
Itekhnoloji yeLiveness eboniswe yiVisionLabs igqithise iziphumo zomthathi-nxaxheba kwindawo yesibini ngamaxesha angama-1,5. Ngamaqela angama-25 aphuma kumazwe ahlukeneyo athathe inxaxheba kwinqanaba lokugqibela lokhuphiswano. Iziphumo zayo zinokufunyanwa .
Imveliso yeflegi yenkampani liqonga lokuqaphela ubuso be-LUNA. Isekelwe kwi-algorithm ye-LUNA SDK, eye yathatha ngokuphindaphindiweyo izikhundla eziphambili kwiimvavanyo ezizimeleyo emhlabeni jikelele. Inkqubo isetyenziswa ziibhanki ezingaphezu kwama-40 kunye namaziko amatyala esizwe eRashiya nakumazwe eCIS.
umthombo: 3dnews.ru