Xa uthathe i-smartphone okanye i-camera kunye nokufumana isithombe eshushu kwi- "auto mode", isithombe esekelwe ngokugqithisileyo, esigqithisileyo kwi-colour, kunye ne-focus kakhulu. Phantsi kwesixesha le-magical, i-Image Signal Processor (ISP) ye-camera kunye ne-core "decision intelligence" yayo - i ‒ Ukusebenza ngempumelelo. I-3A ifumaneka i-Auto Exposure (AE), i-Auto White Balance (AWB), kunye ne-Auto Focus (AF). Ezi ezintathu zihlanganisa njenge-photographers professional e-camera, zihlanganisa ukuxhaswa kwimeko embalwa embalwa. 3A algorithms Kuba ama-imaging engineers, umgangatho we-algorithms ye-3A ikakhulukazi ngokuthe ngqo kwimveliso yomsebenzisi. I-system ye-3A elungileyo ibonelela kubasebenzisi ukujongana kwi-composition kunye ne-emotional expression; ngokufanelekileyo, i-3A engapheliyo ingaba kuholele kwimiphumo ye-exposure, imibala embalwa, okanye iifoto ze-out-of-focus ezininzi. Le nqaku iya kuxolisa kwi-mysteries ye-3A. I-AE (I-Auto Exposure): Indlela yokubala ukukhanya ngokufanelekileyo? I-Auto Exposure (AE) isicelo yokuqala yokulawula i-brightness ye-image, ukuqinisekisa ukuba i-frame ayikho i-overexposed okanye i-underexposed. I-key ikhona kwi-"metering", nto leyo kuquka ukulawula i-intensity ye-light ye-scene kunye nokuguqulwa kwe-apertura, i-shutter speed, okanye i-ISO ngokufanelekileyo. Iimpawu kunye neengxaki ze-Metering Modes Inkqubo yokubala ye-camera isekelwe kwi-assumption esisiseko: ibonelela zonke izicwangciso zibe i-18% ye-reflectance, i-"middle grey." Xa ukunyuka kwe-scene epheleleyo kulinganiswa ne-standard, i-AE inokukwazi ukunika iiparametrithi ezifanelekileyo ze-exposure. Nangona kunjalo, kwihlabathi esemgangathweni. Xa ukwamkela i-snow (i-high-reflectance), i-camera ingaba ngempumelelo ukunciphisa i-exposure, ukuguqula i-snow ye-white ukuya kwi-grey; xa ukwamkela i-cat ye-black (i-low-reflectance), ingaba ukwandisa i-exposure, Ukuze usebenzise iindawo ezininzi, iingcali zenzelwe iindlela ezininzi zokusetyenziswa: 
 
 
 
 I-Matrix/Evaluative Metering: Le mod ye-intelligent and most commonly used. I-Matrix/Evaluative Metering ibandakanya i-frame kwiiyunithi ezininzi, ibandakanya ngokuhambisana i-brightness, i-color, kunye ne-focus point of each zone, kwaye isetyenzise i-database ye-image ye-internal kunye ne-algorithms ezininzi zokusetyenziswa i-balanced exposure value. Le mode isebenza kakuhle kwiimeko ezininzi kunye ne-default camera 'all-rounder. I-Center-Weighted Metering: Le mode ivumela ukufikelela kwindawo ye-central ye-frame (kuxhomekeke kwi-20-30%), nangokuthi ukhangela ubushushu wokugqibela kunye ne-weighted averages. I-Metering iyatholakala kwiimeko apho i-subject ye-central ibekwe kwi-centre, njenge-portraits ezivamile. Spot Metering: I-spot meter ukunyusa umlilo kuphela kwiindawo enobuncinane kakhulu esekelwe kwicandelo (ku-1-5%), ngokugqithiselwe ngokupheleleyo kwiindawo ezininzi. Oku inikeza abaphotographers ukulawula ngokukhawuleza, ikakhulukazi kwizithuba ezininzi ze-contrast ezifana ne-back-it portraits okanye i-stage photography. I-Engineer's Challenge: Ukuphendula i-high-contrast scenes Enye yeengxaki ezininzi kwi-imaging iimveliso yokuvelisa iingxaki ezininzi ze-contrast ezifana ne-backlighting okanye i-sunrise / i-sunsets. Kwiimeko ezininzi, nayiphi na indlela ye-single ye-metering kunokwenzeka. I-AE algorithms ezifanelekileyo isetyenzise iinkqubo ezininzi ezisetyenzisiweyo, njenge: 
 
 
 
 Ukusetyenziswa kwe-scene: Ukusetyenziswa kwe-AI ukuyifumana i-scene njenge- "i-backklit portrait" okanye "i-sunset" kwaye usetyenzisa i-curve ye-exposure. I-Face Priority AE: Xa i-face ifumaneka, ukunika i-priority efanelekileyo ye-face, nangona i-background inokufumana i-overexposure. I-HDR (i-High Dynamic Range) Fusion: Ukukhuthaza ngokuzenzakalelayo izibane ezininzi kwaye zihlanganisa kwi-image enye kunye neengxaki ezininzi kwi-highlights kunye ne-shadows. I-AWB (I-Auto White Balance): Yintoni i-camera inokufunda umbala ye-light? I-Auto White Balance (AWB) yenzelwe ukuguqulwa kwe-colour casting eyenziwe ngama-color temperatures ezahlukileyo ze-light source, yenza izicwangciso ze-white ziyafumaneka kwi-white under any lighting. I-eye yabantu iye i-colour constant, i-automatically adapting to ambient light, kodwa amacandelo ze-camera ayikho; bafumana kuphela i-color ye-light njengoko. Ukusuka kwi-Gray World Assumption ukuya kwi-Machine Learning Ukuze ufumane iingxaki ye-AWB, kufuneka kuqala ukufumana iimodeli ye-physical ye-image formation. I-"Observed Color" eboniswe yi-sensor yi-product ye-ambient "Illuminant Spectrum", i-"Surface Reflectance" ye-object, kunye ne-sensor's own "Sensor Response" curve: Observed Color = Illuminant Spectrum × Surface Reflectance × Sensor Response I-camera inokufumana kuphela i "I-Observed Color", kwaye le nqakraza ye-AWB kuyinto ukufikelela kwi "I-Illuminant Spectrum" ukusuka kwakhona ukuguqulwa. Nangona kunjalo, ngenxa yokuba i-"Surface Reflectance" ayaziwa, oku kubonisa i-classic "i-ill-posed problem" - i-equation ye-multiple unknown that cannot be solved directly. Ngenxa yoko, i-AWB algorithms ezivamile isekelwe kwiingxaki ze-statistics: 
 
 
 I-Gray World Assumption: Yinto yeyona ye-theory ye-classic, ekubeni ukuba kwi-color-rich image, i-average yeentsimbi zehlabathi iya kuxhomekeke kwi-neutral grey. I-algorithm i-calculates the overall R, G, B averages of the frame; ukuba i-average ivela kwi-grey, umzekelo, ukuba i-yellowish, i-system iya kubonelela i-canal blue ukuba yibhalise. I-Perfect Reflector/White Patch Assumption: Le ngxaki ibonisa ukuba i-point eluhlaza kwicandelo kufuneka i-white. I-algorithm ibonelela i-point eluhlaza kwicandelo kwaye ibonelela kwi-white. Nangona kunjalo, ezi zixazululo ezincinci ziyafumaneka. Umzekelo, kwi-green gazon enkulu, i-grey world i-assumption ingaba ngokufanelekileyo ukongeza i-red ukuze "ukuguqula" i-green, okuholela kwi-purple casting. Kwi-illumination ehlanganisiwe (isib. i-Office eyenziwe nge-window kunye ne-daylight kunye ne-fluorescent light), i-algorithms ezivamile ziyafumaneka. I-Engineer's Challenge: I-Tuning under Complex Lighting Ngenxa yeempembelelo yeemodeli yokufakelwa kwangaphambili, i-AWB tuning iyiphi na iinkcukacha ezinzima ze-image engineering. Ukuphathelela iintlobo ezahlukeneyo zangaphakathi zangaphakathi zangaphakathi, njenge-office fluorescent lights, i-home incandescent lights, okanye i-sunrise lights, iingcali zihlanganisa i-database enkulu yeentsimbi zangaphakathi. I-AWB algorithms zangaphakathi zangaphakathi zangaphakathi zangaphakathi, zangaphakathi kwiimodeli zangaphakathi zangaphakathi ze-machine learning: 
 
 
 I-Algorithm ibonise iimpawu ze-colour distribution ye-scene kwaye ibonelela kunye neemodeli ezininzi ze-light source kwi-database yayo (isib. I-D65 daylight, i-A-source incandescent light) ukuze "ngathi" i-corrent ambient light.  Since applying independent white balance correction to different regions of the image (i.e., regional white balance) carries a very high perceptual risk, easily leading to color artifacts or unnatural transitions, actual implementations often adopt the following strategies:
 
 Mixed Lighting Processing: 
 
 
  The algorithm estimates multiple potential light sources in the scene and calculates a "compromise gain" that balances color across all parts of the image. This method aims to maintain overall color continuity and naturalness, avoiding visual flaws caused by local corrections. Global Optimal Solution:  Utilizing AI scene recognition technology, the algorithm can identify key areas in the image, such as faces, skin tones, or specific objects. When calculating the global white balance gain, higher weights are given to these visual focal points to ensure the accuracy of the main subject's colors. Simultaneously, background areas may retain some ambient light atmosphere to achieve visual balance and naturalness. Semantic Weighting: I-AWB efanelekileyo kufuneka isebenze ngokubanzi kuphela phantsi kweengxaki zokusetyenziswa, kodwa kufuneka ziyafumaneka ngexesha lokucacileyo kunye neengxaki ze-monochromatic, ukunciphisa iingxaki ze-colour njengoko iingxaki zihambe. Oku kubalulekile kwimvavanyo elungileyo yabasebenzisi. I-AF (i-Auto Focus): Ukukhangisa phakathi kwe-speed kunye ne-accuracy I-Autofocus (AF) inesibophelela ukuguqulwa kwe-lens ukuze yenza i-objekthi emangalisayo. Ukusuka kwi-early "hunting" autofocus ukuya kwi-"point-and-shoot" precision namhlanje, ubuchwepheshe ye-AF iye yandiswa kakhulu. Ukusuka kwi-Contrast Detection AF ukuya kwi-Phase Detection AF (PDAF) 
 
 
 I-Contrast Detection AF: Umgangatho wayo i-contrast at the focus point is highest when the image is sharpest. I-camera isiguquli ngokugqithisileyo i-lens kuze kufumana i-peak contrast. I-avantage yi-accuracy ephezulu, kodwa i-disadvantage yi-speed ephantsi. Ukusetyenziswa kwe-Phase Detection AF (PDAF): Ukusetyenzisa izakhiwo ezizodwa ze-pixel kwi-sensor (njenge-microlens array okanye i-pixel masked) ukuxhaswa i-light eyenza kwiindawo ezahlukeneyo ze-lens (njenge-left kunye ne-right), ukwakha iifoto ezimbini ezahlukileyo. Ngokusho ukuxhaswa kwe-relative displacement (i-image phase difference) phakathi kweifoto ezimbini ezahlukileyo, i-algorithm inokufuneka ngokukhawuleza ukuba i-focus ibekwe phambi okanye emva kwe-object, kunye ne-distance elidlulayo ukuba i-lens ibekwe ukufumana i-focus. Le nkqubo ivumela ukuba i-lens I-Engineer's Challenge: Ukubhalisa i-Focus Speed kunye ne-Accuracy Kwi-development ye-AF lens drivers, i-challenge yayo yeengcali yokufumana i-balance phakathi kwe-speed kunye ne-accuracy. Nangona i-PDAF i-incredibly fast, i-PDAF inezidingo ezithile ye-light kunye ne-object contrast, kwaye i-DSLR lens ye-traditional ingangena neengxaki ze-focus shift. I-Contrast Detection AF, nangona i-contrast detection, isebenzisa iinkcukacha ngqo kwi-sensor yokuqala, ibonelela i-focus epheleleyo. Ngenxa yoko, iikhamera ezintsha kubasetyenziswa kakhulu "i-Hybrid Auto Focus", enxibelelanisa iintlobo ezimbini: 
 
 
 Isinyathelo sokuqala (I-Adjustment Coarse): Ukusebenzisa i-PDAF ukuhambisa ngokukhawuleza i-lens phantsi kwe-point focal. Isinyathelo yesibini (Fine Adjustment): Ukuguqulwa kwi-Contrast Detection AF ukuze kuqinisekisa ukucaciswa okucacileyo, ukunika ukucaciswa okuhle. Ukulungiselela: Ukusebenzisana umsebenzi 3A ‒ A Millisecond-level Team Effort Okokuqala, kufuneka ibonise ukuba AE, AWB, kunye ne-AF ayikho iimoduli ezahlukileyo, kodwa Ukubonisa i-team ye-photography professional: Xa ngexesha lokucoca i-shutter, i-gaffer (AE), i-colorist (AWB), kunye ne-camera assistant (AF) kufuneka ifumaneke kunye nokuxhomekeke ngexesha elifanelekileyo kwi-milliseconds ukufumana iifoto epheleleyo. Tightly Coupled Closed-loop Control System Kwi-ISP efanelekileyo, i-module ye-3A isebenza njenge-team efanelekileyo: imiphumela ye-statistics yeqela elinye isetyenziswa ngexesha elifanelekileyo kwi-AE, i-AWB, kunye ne-AF modules, ngokucacisa izixazululo ze-frame elandelayo. Oku kuthetha ukuba i-design ye-3A yenzelwe , hhayi iingxaki ezintathu ezahlukileyo. Zifumaneka ngokugqithisileyo ukuphucula umgangatho umgangatho umgangatho umgangatho, ngaphezu kwe-extreme ye-parameter eyodwa. Multi-variable Optimization Problem Ukucaciswa okuphumelela kakhulu kwimeko yobuchwepheshe. Qinisekisa i-scenario apho umqhubi uqhuba kwimeko emangalisayo emangalisayo emangalisayo kwimeko emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo emangalisayo 
 
 
 I-System Oscillation: I-system epheleleyo ye-imaging isibambisane ngexesha lokuphendula, njenge-brightness okanye i-colour ebandakanya ngokushesha kwi-frame, engakwazi ukuvelisa ngempumelelo. I-Convergence ye-slow: Ngenxa yeengxaki ze-module, ngaphandle kwe-mechanism efanelekileyo, inkqubo ingaba ixesha elide ukufumana i-exposure, i-white balance, kunye ne-focus eyenziwe ngempumelelo. Ngenxa yoko, inkqubo ye-top-tier 3A kufuneka akubonise kuphela ukucacisa kunye nokuvelisa kwimoduli ngamnye, kodwa nangokufanelekanga ngokufanelekileyo iintlobo zayo, njengoko iqela lefotographer olufanelekileyo. I-AF ibonise i-subject yokuqala, i-AE ibonise i-illumination ngokufanelekileyo kwi-subject kunye ne-environment, kwaye i-AWB ibonise imibala ebonakalayo kakhulu kwi-scene. Kwiiminyaka ezidlulileyo, iimveliso zilungiselela ukuhlanganisa ubuchwepheshe ye-AI, ukusetyenziswa kwe-deep learning ukunceda i-algorithms ze-3A "ukufumana" ngokugqithisileyo iimpendulo yefotographer Preview: Ukulungiselela ngokubanzi kwi-ISP kunye ne-Image Quality Kule nqakraza, siye kwenziwe indlela i-algorithms ye-3A ibhekwa njenge- "intelligence ye-decision" ye-camera, enomdlaza i-exposure, i-white balance, kunye ne-focus. Nangona kunjalo, ukusuka kwedatha emnyama eyenziwa yi-sensor ukuya kwiifoto ezincinane ezibonakalayo ezibonakalayo ezibonakalayo, zonke ziye zihlanganiswe yi-Image Signal Processor (ISP). Kule nqakraza elandelayo, siya kufumana i-mystery ye-ISP, ukufundisa indlela yokuguqula idatha emnyama kwiifoto ezibonakalayo kunye nokufunda kwiimodoli zayo eziphambili, ezifana ne-Demosaic, Denoise, Sharpening, and Color M

Walkthroughs, tutorials, guides, and tips. This story will teach you how to do something new or how to do something better.

Read My Stories

Le audio iveliswe ngolwimi lokuqala lwebali!

I-An Image Engineer's Notes, I-Part 3: Ngaphandle kwe-Camera 3A "I-Decision Intelligence"

About Author

IZIMVO

ZIJONGE IIMPAWU

ELI NQAKU LINIKEZELWE KU

Related Stories

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps