Uhlalutyo lwevidiyo ludibanisa: yintoni eyenziwa yingqondo kunye noomatshini ngobuso bethu

Ukukwazi ukubona nokuqonda ngokukhawuleza ubuso ngamandla amakhulu. Akukho mfuneko yokuchitha ixesha ekuhlalutyeni, ukufunda imibimbi, ukugoba kunye nama-ovals. Ukuqondwa kobuso ngoko nangoko kwaye akulula. Kulula kakhulu ukuba asiyiqondi indlela esiyenza ngayo.

Cinga ngendlela ubuso obahlukeneyo obufana ngayo omnye komnye - amehlo amabini, umlomo, impumlo, iindlebe eziphuma emacaleni, ixesha ngalinye ngokulandelelana okufanayo (izihlandlo ezininzi). Kuyamangalisa ukuba sihlalutya into ngokulula.

"Sicwangciswe" ukuba siqaphele ubuso ukususela ekuzalweni, kodwa ngoku abantu baye benza ngcono - bafundisa umatshini lo buchule. Ukwaziswa ngokubanzi kweenkqubo zokuqondwa nokuchongwa kwabantu kuya kubuchaphazela njani ubomi boluntu?

I-Pareidolia: ukukhangela ubuso obuzenzekelayo

Abantu abakwimo "ezenzekelayo" bayakwazi ukwahlula imifanekiso eqhelekileyo kuyo nayiphi na indawo. Izinto ezintathu nje zokwakha zesakhiwo zibonwa njengobuso bedada elimangalisayo. Lo ngumzekelo we-pareidolia.

Igama elithi pareidolia livela kumagama esiGrike para (para - kufutshane, ejikeleze, ukutenxa kwinto ethile) kunye ne-eidolon - umfanekiso. Eli ligama lenkohliso yamehlo, imbono yomfanekiso okanye intsingiselo apho ungekhoyo. Umzekelo, ubuso kwisiqu somthi okanye imifanekiso yezilwanyana emafini yi-pareidolia.

Iifoto ezongezelelekileyo ezinje zinokufumaneka apha thingswithfaces.com

Sibona ubuso babantu kunye nobuso bezilwanyana naluphi na umfanekiso wejometri. Yonke inkcubeko ye-emoji yakhiwe kulo mgaqo. 🙂

I-phenomenon ye-pareidolia inokuguqulelwa ngokulula kulwimi lwe-algorithmic. Amagcisa Shinseungback Kimyonghun amafu afotiweyo okomzuzwana adityaniswa ebusweni babantu esebenzisa iscript nethala leencwadi le-OpenCV.

I-Thatcher Illusion: Iimpazamo zeSistim yebhayoloji

Kukho ibug yebhayoloji ebonisayo ukubaluleka okukhulu kwesakhono sokuqaphela. Uninzi lwezinto ezikujikelezile - isitulo, itafile, ikhompyuter - kulula ukuyibona kwaye uchonge ngokuchanekileyo kuyo nayiphi na i-angle. Hayi nje ubuso.

Ubuso obujijekileyo budala ukungasebenzi kakuhle kwengqondo okubizwa ngokuba yi-Thatcher effect (inkohliso). Isenzeko sichaza imeko apho kunzima ukubona utshintsho lwendawo kwifoto ejonge phantsi.

Masijike ifoto kaMargaret Thatcher kwaye sijonge iziphumo.

Ifoto yokuqala ibonakala iqhelekile, kodwa ukuba uyayijika, indawo engalunganga yamehlo kunye nomlomo ngokukhawuleza ibamba iliso lakho. Umntu kunye nenethiwekhi ye-neural eyenziweyo babona imifanekiso ngokwahlukileyo. Kuyamangalisa ukuba "inethiwekhi ye-neural" phakathi kweendlebe zethu kulula kakhulu ukuyikhohlisa.

I-Thatcher Illusion ibonisa ezinye zeendlela ezisisiseko apho ubuchopho bethu busebenza ngolwazi. Ingqondo ifunda ingqokelela yezinto ezizimeleyo: ipere yamehlo, impumlo, umlomo kunye neendlebe. Ukongeza kwiimpawu zomntu ngamnye weempawu zobuso, ubudlelwane babo kunye nomnye kunye nendawo zithathelwa ingqalelo. Oko kukuthi, ubuso bubonwa njengenkqubo edibeneyo.

Ke ngoko, xa siboniswa ubuso obujonge phantsi, kunzima ngakumbi ukuba ingqondo ivavanye umfanekiso uphela - ulwazi "luqokelelwa" ngokwahlukileyo kwinto nganye: amehlo asendaweni, umlomo ujongeka njengomlomo. Nangona kunjalo, ngokukhawuleza xa siboniswa ubuso obuchanekileyo, imbono yenkqubo enye idibanisa ngokukhawuleza kwaye iingxaki ziqala: kuyacaca ukuba iimpawu eziqhelekileyo zidibene ngendlela engavamile.

Kutheni ibalulekile nje? Ingqondo yomntu iyakwazi ukuqaphela umahluko omncinci kwiimpawu zobuso ngenxa yengqibelelo yokuqonda. Indawo ye-cerebral cortex ibona ubuso kwaye imisela isalathiso sokujongwa, i-amygdala kunye ne-insula zihlalutya imbonakalo yobuso, kunye nendawo ekwindawo yangaphambili yelobe yangaphambili kunye nenkqubo yolonwabo yobuchopho ivavanya ubuhle bayo.

I-bug njengenqaku: ubuso bukaChernov

(с)

Uphawu lombono womntu lusetyenziselwa ukuhlalutya idatha ehlanganisiweyo ye-multidimensional usebenzisa "ubuso". Ingcali yezibalo yaseMelika uHerman Chernov ngowe-1973 yachaza ingcamango yokusebenzisa “ubuso” ukuchonga ukuxhomekeka kweempawu kunye nokuphonononga ubudlelwane obunzima phakathi kwezinto ezininzi.

Idatha ye-Chernov ibonakaliswe kwi-pictogram yobuso, apho ixabiso elihambelanayo lezinto ezikhethiweyo ezikhethiweyo zibonakaliswa njengemilo kunye nobukhulu beempawu zomntu ngamnye: ubude bempumlo, i-angle phakathi kwamashiya, ububanzi bobuso - ukuya kuthi ga kwi-36 eguquguqukayo iyonke. Ngaloo ndlela, umbonisi unokubona iimpawu ezibonakalayo zezinto ezikhethekileyo kuqwalaselo ngalunye lwamaxabiso.

Ukukhangela ngokukhawuleza kumzobo owenziwe ngobuso kuya kukuvumela ukuba unqume ngokukhawuleza ukuba iimpawu zeeprofayili zihluke kakhulu (ngokuzenzekelayo). Ngokuphononongwa ngokucokisekileyo kweempawu zobuso, kuyacaca ukuba zeziphi iimpawu (inkalo nganye yobuso yinto eyahlukileyo yesethi yedatha yokuqala) zifana kwaye zihluke. Umzekelo, kulo mzekeliso ungasentla, kulula ukubona umahluko phakathi kwamazwe ngokweemvakalelo ezilusizi nezolonwabo.

Kutheni imoto ifuna ubuso bakho?

Ubuchule bokuqaphela ubuso ngokukhawuleza bukunceda ukuba uthabathe umntwana wakho kwi-kindergarten, ukhethe iqabane, kwaye ubonise iimvakalelo ngokuchanekileyo nangokufanelekileyo. Kodwa kwenzeka ntoni xa umntu edlulisela obu buchule kwinethiwekhi ye-neural eyenziweyo?

Ingcamango inokuba lizothe. Ayinguye wonke umntu olungele ukwamkela ngokulula itekhnoloji egcina idatha, ibeke iliso kwiintshukumo, ihlalutye ukuthengwa kunye neemvakalelo. Utshintsho olusuka kuphononongo olulula lwevidiyo ukuya kuhlahlelo lwevidiyo olulolwakho lubandakanya ukonyuka koxanduva.

Namhlanje algorithms ezifana I-DeepFace Fumana ukufana kobuso ngokuchaneka okungaphezulu kunokwabantu. I-algorithm yeNvidia idala ubuso babantu abangekhoyo kwimizuzwana embalwa nje. Ubuso kwikholaji engentla yenziwe I-StyleGAN inethiwekhi ye-neural, eqeqeshwe kwisethi yedatha ye-70 yemifanekiso. Bakhangeleka ngokwenyani ngendlela eyoyikisayo.

Umboniso weSearchFace algorithm

Ekuqaleni, i-algorithm yokuqaphela ubuso buka-Facebook yabangela ukwanda kokuphaphama, kodwa ke wonke umntu waqhelana nayo (okanye wayishiya inethiwekhi yoluntu). Inkonzo ye-FindFace yokukhangela abantu ngesithombe kwi-VKontakte ifumene iimpendulo ezixubileyo kwaye yayisetyenziselwa ukuxhatshazwa, kodwa ukuvalwa kweprojekthi efanayo ye-SearchFace sele ibangele ukusabela okungalunganga kubasebenzisi - emva kwayo yonke into, ukuba idatha ikhona, ke mayivuleleke. kumntu wonke.

Amatyathanga okuthengisa afaka iteknoloji yokuqaphela ubuso ukuthintela ubusela, ukuqokelela idatha malunga nobudala, isini kunye neemvakalelo zabathengi. Ekupheleni kosuku, injongo kukuphucula amava omthengi kunye nokwenza imali kuyo. Xa abathengi beqaphela ukuba inkqubo inenzuzo kubo ngokwabo, abaninzi baya kuvuma ukuphumeza iteknoloji entsha.

Xa iimeko “zobusela besazisi”—ikhadi letyala kunye nobuqhophololo besazisi—esanda, abathengi baya kukhetha inkqubo ekhoyo xa beyifuna. zichonga ngokuchanekileyo.

Okwangoku, i-algorithms inceda ukuxazulula iingxaki zezakhelo zokukhanya okuphantsi, isisombululo esiphantsi kunye ne-camouflage, njengeeglasi, iiwigi kunye ne-multi-day stubble. Iinkqubo zisebenza ngesantya esimangalisayo kwaye zihambelana nobuso kwidathabheyisi yezigidi zabantu ngomzuzwana nje.

Ezinye iivenkile e-USA nikela Abarhanelwa bobusela banokuzikhethela: bazivumele ukuba bafotwe okanye babekwe ityala ngokusesikweni ngolwaphulo-mthetho. Isela lifumana inkululeko kunye nokuvinjelwa kokutyelela ivenkile, kwaye ifoto yakhe ifakwe ngokusemthethweni kwisiseko sedatha. Iifayile ezinemifanekiso yabantu zifihliwe kwaye zifikeleleka kuphela kumnini wenkqubo.

Ngubani owenza ingeniso ngokuqatshelwa?

Uninzi lweevenkile sele zifakile iikhamera zeCCTV. Uhlalutyo lwevidiyo, uhlaziyo lwehardware alufunwa - vele uqhagamshele kwinkonzo yelifu. Kwaye ngenkonzo yohlalutyo lwevidiyo ye-Ivideon, akukho sithintelo sokungena. Iindleko zesisombululo ukusuka kwi-ruble ye-1 ngekhamera nganye inika nawuphi na urhwebo ukufikelela kwisoftware.

Eyona njongo iphambili yokuba abathengisi basebenzise iteknoloji yokuqaphela ubuso kukuthintela ubusela. Ngoku ka inikiwe Ngokutsho kweNational Retail Foundation, e-US kuphela, malunga ne-1,33% yazo zonke iimpahla kwi-2017 zalahleka ngenxa yobusela - akukho ngaphantsi kwe-46,8 yeebhiliyoni zeedola kumonakalo.

Ubugcisa bokuqaphela ubuso bunciphisa ubusela evenkileni ngaphezu kwe-30%.

Ngokuqhelekileyo inani lomonakalo liphenjelelwa zizinto zesibini: ukungakhathali kwabasebenzi, uqeqesho olubi lwenkonzo yokhuseleko, umnqweno wokulondoloza imali. Ezi kunye nezinye iingxaki kufuneka zisonjululwe kusetyenziswa iikhamera kunye nobuchwepheshe bamafu.

Inkqubo yokuqwalasela ubuso iququzelela umsebenzi okhawulezayo kunye noluhlu "olumnyama": luthelekisa ifoto yomxhasi kunye nedatha yobuso obungathembekiyo kwaye, ukuba kukho umdlalo, ithumela isilumkiso esihambelanayo kubagcini bokhuseleko.

Isoftware yokuhlalutya iphucula kakhulu ukhuseleko lwevenkile. Isela elinamava liyakwazi ukubona “iindawo ezingaboniyo” kwiikhamera. Kule meko, umlindi unokusebenzisa ifowuni yakhe ukuthatha ifoto yomrhanelwa aze ajonge ukuba loo mntu ukho kwi-database.

Iimveliso sele zisebenzisa intengiso yeselula ixesha elide - ukuthumela i-SMS, izaziso zokutyhala kunye nokubonisa intengiso ekujoliswe kuyo. Kwiintengiso zemveli, iinkqubo zokuqaphela zibonelela ngamathuba afanayo nabathengisi be-intanethi abawafumene ngeekuki.

Iqonga elifanayo elisetyenziselwa ukuchonga amasela linceda abathengisi ukuba babone ukuba zeziphi iivenkile ezingcono ekutsaleni abathengi. Inkqubo yokuqaphela inceda ukuchonga umxhasi weVIP ekunene emnyango wevenkile. Ukusebenzisa idatha evela kwi-CRM, umthengisi unokukhawuleza enze umnikelo onenzuzo kumxhasi.

KwiZiko lezeMali laMazwe ngaMazwe laseSeoul, iikhamera zolwazi zimi ngexesha langempela chaza ubudala kunye nesini somntu, kwaye unikeze intengiso ngokweeparamitha ezichongiweyo

Ulwazi lwabathengi lusebenzisa isixhobo esinamandla sokunyusa intengiso kunye nokuvavanya iimfuno zabaphulaphuli. Iikhamera ziya kukunceda wenze umboniso weentengiso zevidiyo kundwendwe oluthile ngokuxhomekeke kwisini sakhe, ubudala kunye nemeko yeemvakalelo, kwaye ziya kuba ngababoneleli bedatha ekubaleni ukusebenza kwentengiso.

Amathuba angentla abathengisi ahlala evakala njengengxolo yentengiso ecaphukisayo. Ezi malunga "nokukhula kwenzuzo" kunye "neemfuno zabaphulaphuli" zikhapha nasiphi na isixhobo se-IT kwimarike - ukusuka kwi-ERP ukuya kwiithegi zexabiso le-elektroniki. Ngaba kukho okungakumbi kwiinkqubo zokuqaphela ubuso kunokuthengisa okucocekileyo malunga nobukrelekrele bokwenziwa kunye nobuchwepheshe bexesha elizayo? Masiphendule lo mbuzo ngemizekelo yokusebenzisa iinkqubo zangempela kwiivenkile ezikhoyo.

"Sebenza entsimini": ngubani obona ubuso kwiimeko zokwenyani

I-7-Eleven yeyona nto ithengisa kakhulu kwihlabathi, isebenza ngaphezu kwe-36 yeevenkile ezincinci kumazwe angama-000 phantsi kolawulo lwe-Seven-Eleven Japan. Kutshanje inkampani ifakiwe isoftware kwiivenkile zayo eziyi-11. Itekhnoloji yokuqaphela ubuso kunye nohlalutyo lokuziphatha kwinethiwekhi yokuthengisa isetyenziselwa ukuchonga abanini bamakhadi okunyaniseka, ukubeka iliso kwi-traffic yabathengi, kunye nokumisela inqanaba loluhlu lweempahla kwiindawo zokugcina iimpahla.

I-Saks likhonkco lokuthengisa iprimiyamu yenkulungwane yakudala ephantsi kweyeyona nkampani indala ehlabathini (eyasekwa ngo-1670), i-Hudson's Bay Company. Uhlalutyo lwevidiyo iyasetyenziswa eSaks ikakhulu ukuthintela ubusela. Le software ijonga iifoto zabantu abarhanelwa ukuba ngamasela evenkileni ijonga kuvimba weenkcukacha wamasela aziwayo. Iikhamera zinothungelwano ukuze iziphumo zijongwe kwikomkhulu laseSaks eNew York.

Ngokutsho kwe-Guardian, iivenkile zeprimiyamu kunye neehotele zaseYurophu zisebenzisa itekhnoloji yokuqaphela ubuso ukulandelela ii-VIP kunye nabantu abadumileyo ukuqinisekisa ukuba bafumana amava angcono kakhulu.

E-USA, ikhonkco le-burger CaliBurger isebenzisa iteknoloji yokuqaphela ubuso kwiprogram yokunyaniseka. I-kiosk esebenzisanayo "iqaphela" abathengi, ikhumbula ii-odolo kwaye inika izitya ezizithandayo, kwaye yamkele iintlawulo kunye nokuchongwa kobuso.

Inkqubo isusa umqobo wokungena kwinkqubo yokuvuza kubantu abadala abanokuba nobunzima bokusebenzisa ii-apps eziphathwayo, amanqaku okuvuza kunye namakhadi okuthenga ngetyala.

Iinkqubo zokuqaphela ubuso zisetyenziswa kakhulu e-Asiya, ngakumbi e-China, apho zisetyenziselwa ukuhlawula ukutya, ukukhupha imali kwi-ATM, okanye ukuthabatha imali mboleko. Ukuchaneka kokubonwa kobuso eChina kudlula amandla eliso lomntu. Oku kukwangenxa yenguqu enkulu yaseTshayina ukusuka kwi-2D ukuya kwi-3D yokuqaphela.

Kwimeko yokuqala, ii-algorithms zisebenzisa imifanekiso enemigangatho emibini eqokelelwe kwiziko ledatha ukuhlalutya. Ukuqatshelwa kwe-3D kuhlalutya imifanekiso ye-XNUMXD eyakhiwe ngokutsha kwaye ibonisa ukuchaneka okuphezulu kakhulu. E-China, ukuskena ubuso kungasetyenziselwa ukuthenga (umzekelo, ukuhlawula ii-odolo kwi-KFC), ukwenza iintlawulo kunye nokungena kwizakhiwo.

Ku-Alipay kufuneka ndincume, ukwenzela ukuba inkqubo yokuqatshelwa kwentlawulo iqonde: phambi kwayo ayikho ifoto, kodwa umntu ophilayo. Kuthiwa akunakwenzeka ukukhohlisa u-Alipay: ukutshintsha umbala weenwele, izimonyo, okanye ukusebenzisa iwigi ayitshintshi nto. Inkqubo isebenzisa isethi yeempawu ezahlukileyo ezithatha ingqalelo kwijometri yobuso kunye nendawo yamanqaku athile kuyo.

ezifunyanisiweyo

Ubungakanani botyalo-mali oluthe ngqo kwiinkampani zaseNtshona kunye neTshayina kubuchwepheshe bokwamkelwa kobuso mkhulu kakhulu. Nangona kunjalo, eRashiya ukuphunyezwa kweeprojekthi ezinjalo ngumcimbi wexesha. Iinkampani ezinkulu zorhwebo sele ziqonda izibonelelo kunye nefuthe loqoqosho. Ukuba siqwalasela ukubonwa kobuso njengemveliso, kubalulekile ukuqonda ukuba icandelo ngalinye leshishini lineenkcukacha zalo, kubandakanywa namaxabiso. Xa likhulu ishishini, kokukhona iikhamera ezininzi kunye neemodyuli zohlalutyo zinokufuneka. Izisombululo zamashishini amakhulu zihlala zintsonkothile iiprojekthi ezilungiselelweyo, kwaye ukwenziwa ngokwezifiso kufuna imali eyongezelelweyo. Amashishini aphakathi kunye amancinci angakwazi ukufikelela ngokulula ngekhamera enye enemodyuli edibeneyo yokuqaphela ubuso. Kule meko, iindleko zesisombululo zifaniswa nokusebenzisa i-cloud video surveillance.

umthombo: www.habr.com