Ubuchwepheshe Bokucubungula Izithombe
Ukucubungula izithombe kuyinsimu yesayensi nobunjiniyela egxile endleleni amakhompyutha aguqula ngayo, athuthukise, ahlaziye, futhi aqonde ngayo izithombe zedijithali. Empilweni yanamuhla, lobu buchwepheshe bukhona cishe yonke indawo: kusukela kumakhamera eselula athuthukisa izithombe ngokuzenzakalelayo, izinhlelo zokuphepha ezibona ubuso, izinhlelo zezokwelapha ezifunda izithombe ze-X-ray, kuya ezimotweni ezizimele "ezibona" umgwaqo. Ukucubungula izithombe kuyibhuloho elibalulekile phakathi kwezwe elibonakalayo nezinhlelo zekhompyutha, njengoba kwenza ulwazi ngesimo sezithombe luguqulwe lube idatha engacutshungulwa ukuze kukhiqizwe izinqumo.
Incazelo kanye nobubanzi
Ngokuvamile, ukucubungula izithombe kuyinqubo yokulawula isithombe sedijithali ukuze kuthuthukiswe ikhwalithi yaso noma kukhishwe ulwazi oluthile. Izithombe zedijithali zakhiwe ngamaphikseli, amachashazi amancane anobukhulu noma amanani ombala. Ezithombeni ezimpunga, iphikseli ngayinye ngokuvamile inenani eliphakathi kuka-0–255 (omnyama kuya komhlophe). Ezithombeni ezinemibala, amaphikseli ngokuvamile amelelwa ngeziteshi ze-RGB (Obomvu, Oluhlaza, Oluhlaza okwesibhakabhaka), ngayinye inobubanzi obuthile bamanani.
Ububanzi bokucubungula izithombe buhlanganisa imisebenzi eminingana eyinhloko: ukutholwa kwezithombe (ukuthwebula izithombe kusuka kukhamera noma inzwa), ukucubungula kusengaphambili (ukuthuthukiswa kwekhwalithi njengokunciphisa umsindo), ukuguqulwa (isibonelo ukujikeleza noma ukwandisa izithombe), ukuhlaziywa (njengokuthola izinto), kuya ekuchazeni okuvame ukuhlotshaniswa nombono wekhompyutha kanye nobuhlakani bokwenziwa.
Izigaba Eziyisisekelo Zokucubungula Izithombe
Empeleni, ukucubungula izithombe kuvame ukulandela ukulandelana kwezinyathelo okuhlelekile. Okokuqala ukuthola izithombe, inqubo yokuthola izithombe kusetshenziswa amadivayisi anjengekhamera yedijithali, ama-scanner, ama-drones, noma izinzwa zezokwelapha ezifana ne-CT scans nama-MRI. Ikhwalithi yokuthola izithombe ibalulekile emphumeleni wokugcina, njengoba izithombe ezimnyama kakhulu, ezifiphele, noma ezinomsindo kunzima ukuzihlaziya.
Isigaba sesibili ukucubungula kusengaphambili. Umgomo ukuthuthukisa ikhwalithi yesithombe ukuze kube lula ukuyisebenzisa ezigabeni ezilandelayo. Izibonelo zifaka phakathi ukunciphisa umsindo (ukuphazamiseka kwezinhlamvu noma amachashazi), ukulola, ukulinganisa ukukhanya, noma ukulungiswa kombala. Ukucubungula kusengaphambili kungafaka futhi ukwenziwa kwesayizi wesithombe kube ngokwejwayelekile kanye nokuguqulwa kwefomethi, isibonelo, kusuka ku-RGB kuya ku-grayscale ukuze kube lula ukubala.
Isigaba sesithathu ukuhlukaniswa, okuhlukanisa izindawo ezibalulekile nesizinda. Ukuhlukaniswa kuyadingeka ezindleleni ezahlukene, njengokuhlukanisa amaseli egazi nesizinda esithombeni se-microscope, noma ukuhlukanisa izimoto emgwaqweni nezindawo ezizungezile. Izindlela zokuhlukanisa zingaba lula njengokunquma imingcele (ukunquma imingcele yamanani okukhanya nobumnyama) noma kube nzima njengokuhlukaniswa okusekelwe ekufundeni okujulile.
Isigaba esilandelayo ukukhipha izici, okuhlanganisa ukukhipha izici ezithile esithombeni, njengemiphetho, ukuthungwa, ukuma, noma iphethini. Lezi zici zisetshenziselwa ukubona izinto noma ukuhlukanisa izithombe. Okokugcina, ukuqashelwa nokuchazwa, njengokunquma ukuthi isithombe siqukethe isimila, ukuthi ubuso buhambisana nedatha ethile, noma ukuthi umkhiqizo unephutha, kuyabandakanyeka.
Amasu Adumile Ekucubungulweni Kwezithombe
Kusetshenziswa amasu ahlukahlukene ekucubungulweni kwezithombe, kuye ngenhloso kanye nohlobo lwesithombe. Enye indlela eyisisekelo ukuhlunga. Izihlungi zingasetshenziswa ukushelela isithombe ukuze kuncishiswe umsindo, noma ukugcizelela imininingwane (ukulola). Isibonelo sesihlungi sokushelela yi-Gaussian blur, kuyilapho izihlungi zokulola zivame ukusebenzisa ama-kernels athile njenge-Laplacian.
Enye indlela ebalulekile ukuthola imiphetho, isibonelo kusetshenziswa opharetha be-Sobel, Prewitt, noma i-Canny. Ukuthola imiphetho kuyasiza ekuboneni imingcele yezinto, ngaleyo ndlela kusiza ukuhlukaniswa kanye nokuhlaziywa kwesimo. Ngaphezu kwalokho, ukuguqulwa kwesimo njengokuguguleka kanye nokwanda kuvame ukusetshenziswa ezithombeni ezimbili ukuze kuthuthukiswe isimo sento, kuvalwe imigodi emincane, noma kususwe umsindo.
Ekucutshungulweni kwezithombe kwanamuhla, ukufunda komshini, ikakhulukazi ukufunda okujulile, kudlala indima ebalulekile. Amanethiwekhi e-Convolutional neural (ama-CNN) ayakwazi ukufunda izici zesithombe ngokuzenzakalelayo ngaphandle kwesidingo sokuklama izici ngesandla. Lobu buchwepheshe busebenze kahle ekuhlukaniseni izithombe, ekutholeni izinto (isb., i-YOLO kanye ne-Faster R-CNN), kanye nokuhlukaniswa kwe-semantic (isb., i-U-Net kanye ne-Mask R-CNN). Ukufunda okujulile kwenza amakhompyutha akwazi ukubona amaphethini okubona ayinkimbinkimbi, ngisho nangaphansi kwezimo zokukhanya ezahlukahlukene kanye nama-engeli okubuka.
Izicelo Ezindaweni Ezihlukahlukene
Ukucubungula izithombe kunezinhlelo zokusebenza ezibanzi emikhakheni eminingi. Kwezokunakekelwa kwempilo, izithombe zezokwelapha zisetshenziselwa ukusiza odokotela ukuxilonga izifo. Izibonelo zifaka phakathi ukuhlaziywa kwe-X-ray ukuthola izifo zamaphaphu, ukucubungula i-MRI ukubona ngeso lengqondo izicubu zobuchopho, kanye nokutholwa komdlavuza kusenesikhathi ngezithombe ze-mammography. Ukucubungula izithombe kungathuthukisa ukungafani, kugqamise izindawo ezisolisayo, futhi kusize ekulinganisweni kosayizi wesimila okunembe kakhudlwana.
Emkhakheni wezokuphepha, ubuchwepheshe bokubona ubuso kanye nokutholwa kwemisebenzi esolisayo kuncike kakhulu ekucutshungulweni kwezithombe kanye nokubona kwekhompyutha. Amakhamera e-CCTV esimanje angahlanganiswa nama-algorithms ukuze aqaphele amapuleti elayisensi yezimoto (i-ANPR), aqaphe izixuku, noma athole ukwephulwa kwemithetho ezindaweni ezithile.
Embonini nasekukhiqizeni, ukucubungula izithombe kusetshenziselwa ukuhlolwa kwekhwalithi. Izinhlelo zokuhlola umbono zingabona imikhiqizo enephutha emigqeni yokukhiqiza, zibale izinto, zifunde amakhodi e-QR noma amabhakhodi, futhi ziqinisekise ukuthi osayizi bezingxenye bahlangabezana nezindinganiso. Izinzuzo ijubane kanye nokuvumelana, njengoba izinhlelo ezizenzakalelayo zingasebenza ngokungakhathali futhi zinciphise amaphutha abantu.
Emkhakheni wezolimo, izithombe ezivela kuma-drone noma amasathelayithi zingahlaziywa ukuze kuqashwe impilo yezitshalo, kutholakale ukuntuleka kwamanzi, kudwetshwe izindawo ezigcwele izinambuzane, futhi kulinganiswe isivuno sezitshalo. Amasu okucubungula izithombe avumela ukuhlaziywa kwezitshalo ngezinkomba ezifana ne-Normalized Difference Vegetation Index (NDVI), esiza abalimi ukwenza izinqumo ezisekelwe kudatha.
Ngisho nasemhlabeni wezokuzijabulisa kanye nezinkundla zokuxhumana, ukucutshungulwa kwezithombe kuvelele. Imiphumela yokuhlunga, ukushelela ubuso, ukushintshwa kwesizinda, ngisho nokudalwa kokuqukethwe kwe-augmented reality (AR) kusebenzisa amasu afana nokuhlukaniswa kwesikhathi sangempela kanye nokulandelela ubuso. Lobu buchwepheshe benza ukusebenzisana kwedijithali kube okuhehayo futhi kube ngokomuntu siqu.
Izinselele Nezinkinga Zokuziphatha
Naphezu kwezinzuzo zayo ezinkulu, ukucubungula izithombe nakho kubhekene nezinselele ezinkulu. Enye ikhwalithi yedatha elinganiselwe: izithombe ezifiphele, ukukhanya okungekuhle, ukucaca okuphansi, noma umsindo kungonakalisa ukunemba kohlelo. Ngaphezu kwalokho, ukufunda okujulile kudinga umzamo omkhulu wokubala, ikakhulukazi ekucubungulweni kwesikhathi sangempela, njengasezimotweni ezizisebenzela zona.
Ngokombono wokuziphatha, inkinga evame ukuphakanyiswa ubumfihlo. Ukusetshenziswa kwamakhamera nokubona ubuso kungaphakamisa ukukhathazeka mayelana nokusetshenziswa kabi kwedatha kanye nokuqapha ngokweqile. Kukhona futhi nenkinga yokubandlulula kwe-algorithmic, njengezinhlelo zokubona ubuso ezisebenza kabi emaqenjini athile ngenxa yedatha yokuqeqesha engalingani. Ngakho-ke, ukuthuthukiswa kobuchwepheshe bokucubungula izithombe kudinga ukulawulwa, ukucaca, kanye nokwakhiwa kohlelo olunomthwalo wemfanelo.
Ikusasa Lokucubungula Izithombe
Ikusasa lobuchwepheshe bokucubungula izithombe kulindeleke ukuthi lithuthuke kakhulu futhi lihlanganiswe. Izinhlelo zizoqhubeka nokusebenza kumadivayisi asemaphethelweni afana nama-smartphone namakhamera ahlakaniphile, okwenza kube lula ukucubungula ngokushesha ngaphandle kokuthembela njalo efwini. Intuthuko kumamodeli e-AI akhiqizayo ivula namathuba amasha, njengokubuyisela izithombe ezindala, ukuthuthukisa isixazululo (isisombululo esiphezulu), ngisho nokudala okuqukethwe okuzenzakalelayo okubonakalayo.
Ngaphezu kwalokho, ukubambisana phakathi kokucubungula izithombe, i-Internet of Things (IoT), kanye ne-robotics kuzosheshisa ukuzenzekela kuzo zonke izinkambu eziningi. Ngama-sensor angcono kanye nama-algorithms ahambisana nezimo eziningi, amakhompyutha azokwazi "ukubona" umhlaba ngokunemba nokuqina okwandayo.
Isiphetho
Ubuchwepheshe bokucubungula izithombe buyisisekelo esibalulekile sezinto eziningi ezintsha zedijithali, kusukela ekunakekelweni kwezempilo kanye nokuphepha kuya embonini kanye nezolimo ngisho nokuzijabulisa. Ngokusebenzisa amasu okucubungula ayisisekelo, ama-algorithms okuhlaziya, kanye nobuhlakani bokwenziwa, izithombe zedijithali zingaguqulwa zibe ulwazi olunenjongo. Kodwa-ke, le ntuthuko kumele ilinganiswe ngokunaka izinselele zobuchwepheshe nezokuziphatha, ikakhulukazi lezo ezihlobene nobumfihlo kanye nokubandlulula. Ngokuthuthuka okunesibopho, ukucubungula izithombe kuzoqhubeka nokuba ubuchwepheshe obuyinhloko obubumba indlela abantu abasebenzisana ngayo nomhlaba obonakalayo enkathini yanamuhla.
