I-VisionLabs, eyingxenye ye-ecosystem ye-Sberbank, iphume phambili okwesibili ekuhloleni ama-algorithms okubona ubuso e-US National Institute of Standards and Technology (NIST).
Ubuchwepheshe beVisionLabs bawine indawo yokuqala esigabeni seMugshot bangena kwabangu-3 abahamba phambili esigabeni seVisa. Ngokuya ngesivinini sokuqashelwa, i-algorithm yayo ishesha ngokuphindwe kabili kunezixazululo ezifanayo zabanye ababambiqhaza. Ngesikhathi somncintiswano, kwahlolwa ama-algorithms angaphezu kwe-100 avela kubahlinzeki abahlukahlukene.
I-NIST yethule ukuhlolwa okusha kobuchwepheshe bokubona ubuso ngoFebhuwari 2017. Ukuhlola kwe-FRVT 1:1 kuhambisana nesimo sokuqinisekisa ubunikazi bomuntu ngokuqinisekiswa kwesithombe. Ucwaningo, ikakhulukazi, lusiza Umnyango Wezohwebo wase-US ukuhlonza abahlinzeki bezixazululo abahamba phambili emhlabeni kule ngxenye yesofthiwe.
Esigabeni se-Mugshot (isithombe sesigebengu, lapho ukukhanya nengemuva kuyaguquguquka futhi nekhwalithi yesithombe ingase ibe mbi), ukuqashelwa kobuso kuhlolwa kusizindalwazi sezithombe zabantu ezingaphezu kwesigidi. Iqukethe izithombe zomuntu ofanayo onomehluko omkhulu weminyaka, okwandisa ubunkimbinkimbi bomsebenzi.
I-algorithm ye-VisionLabs ibona kahle u-99,6% ngenani elingelona iqiniso elingu-0,001%, elingaphezu kwemiphumela yabanye ababambiqhaza. Ukuhlolwa okuhlukile kwahlongozwa kulesi sigaba, okuhlinzeka ngokubona abantu ezithombeni ezithathwe ngokuhlukana kweminyaka engu-14. Kulokhu kuhlolwa, i-VisionLabs ithathe indawo yokuqala (99,5% enenani elingelona iqiniso elingu-0,001%) kuphela, lizihlukanisa njenge-algorithm yokubona ubuso ekwazi ukumelana nobudala.
Esigabeni se-Visa (izithombe zesitudiyo ekukhanyeni okuhle kungemuva elimhlophe), ukuqashelwa kwenzeka ngokusekelwe kudathabheyisi yezithombe eziyizinkulungwane ezingamakhulu ambalwa zabantu. Ubunzima lapha kwakuwukuthi i-database iqukethe izithombe zabantu abavela emazweni angaphezu kwe-100. Kulokhu, i-algorithm ye-VisionLabs ibona kahle u-99,5% ngenani elingelona iqiniso elingu-0,0001%, likleliswe endaweni yesibili kubo bonke abathengisi.
Ngo-April 2019, i-VisionLabs yathatha indawo yokuqala ezigabeni ze-Mugshot, futhi yaba phakathi kwabathathu abaphezulu esigabeni se-Visa.
NgoMashi 2019, i-VisionLabs yathatha indawo yokuqala emqhudelwaneni omkhulu wamazwe omhlaba i-ChaLearn Face Anti-spoofing Attack Detection Challenge engqungqutheleni ye-CVPR 2019, umcimbi omkhulu waminyaka yonke embonweni wekhompyutha.
Ubuchwepheshe be-Liveness obethulwa yi-VisionLabs budlule imiphumela yombambiqhaza ophume endaweni yesibili izikhathi ezingu-1,5. Amaqembu angu-25 avela emazweni ahlukene abambe iqhaza esigabeni sokugcina salo mqhudelwano. Imiphumela yayo ingatholakala .
Umkhiqizo ohamba phambili wenkampani yinkundla ye-LUNA yokuqaphela ubuso. Isekelwe ku-algorithm ye-LUNA SDK, esithathe ngokuphindaphindiwe izikhundla eziholayo ezivivinyweni eziningi ezizimele emhlabeni jikelele. Lolu hlelo lusetshenziswa amabhange angaphezu kuka-40 kanye nezikhungo zezikweletu zikazwelonke eRussia nasemazweni e-CIS.
Source: 3dnews.ru