Kashi na biyu:
Duk wani raster изображение za a iya wakilta a cikin tsari matrix mai girma biyu. Lokacin da yazo ga launuka, ana iya haɓaka ra'ayin ta hanyar kallon hoto azaman matrix mai girma uku, wanda ake amfani da ƙarin girma don adana bayanai don kowane launi.
Idan muka yi la'akari da launi na ƙarshe a matsayin haɗin abin da ake kira. launuka na farko (ja, kore da shuɗi), a cikin matrix ɗin mu mai girma uku mun ayyana jirage uku: na farko don ja, na biyu don kore kuma na ƙarshe don shuɗi.
Za mu kira kowane batu a cikin wannan matrix pixel (kayan hoto). Kowane pixel ya ƙunshi bayani game da ƙarfin (yawanci a matsayin ƙimar lamba) na kowane launi. Misali, pixel ja yana nufin cewa ya ƙunshi 0 kore, 0 shuɗi da matsakaicin ja. Hoton pixel za a iya kafa ta amfani da haɗin launuka uku. Amfani da kewayon lamba daga 0 zuwa 255, ana ayyana pixel mai ruwan hoda azaman Ja = 255, Green = 192 и Blue = 203.
An buga wannan labarin tare da tallafin EDISON.Muna tasowa , sannan kuma munyi alkawari .
Madadin hanyoyin don ɓoye hoton launi
Akwai wasu samfura da yawa don wakiltar launuka waɗanda ke yin hoto. Misali, zaku iya amfani da palette mai maƙasudi, wanda ke buƙatar byte ɗaya kawai don wakiltar kowane pixel, maimakon ukun da ake buƙata lokacin amfani da ƙirar RGB. A cikin irin wannan samfurin, yana yiwuwa a yi amfani da matrix na 2D maimakon matrix na 3D don wakiltar kowane launi. Wannan yana adana ƙwaƙwalwar ajiya, amma yana ba da ƙaramin gamut launi.
RGB
Misali, kalli wannan hoton da ke kasa. Fuskar farko an fenti gaba daya. Sauran su ne ja, koren da kuma shudi na jiragen sama (ana nuna tsananin launuka masu dacewa a cikin launin toka).
Mun ga cewa inuwar ja a cikin asali za su kasance a wurare guda inda aka lura da mafi kyawun sassa na fuska na biyu. Yayin da gudummawar blue za a iya gani kawai a idanun Mario (fuska ta ƙarshe) da kuma abubuwan tufafinsa. Yi la'akari da inda duk jirage masu launi guda uku ke ba da gudummawa mafi ƙanƙanta (ɓangarorin mafi duhu na hotuna) - gashin baki na Mario.
Don adana ƙarfin kowane launi, ana buƙatar takamaiman adadin ragowa - ana kiran wannan adadin zurfin zurfafa. Bari mu ce ana kashe rago 8 (dangane da ƙima daga 0 zuwa 255) kowane jirgi mai launi. Sannan muna da zurfin launi na 24-bit (8 ragowa * 3 R/G/B jiragen sama).
Wani dukiyar hoto ita ce ƙuduri, wanda shine adadin pixels a cikin girma ɗaya. Sau da yawa ana nunawa azaman nisa × tsawo, kamar yadda yake a cikin hoton misalin 4 ta 4 da ke ƙasa.
Wata dukiya da muke hulɗa da ita lokacin aiki tare da hotuna / bidiyo shine rabon al'amari, yana siffanta alakar daidaituwa ta al'ada tsakanin faɗi da tsayin hoto ko pixel.
Lokacin da suka ce wani fim ko hoto yana da girman 16 ta 9, yawanci suna nufin rabon nuni (DAR - daga Nuni Yanayin Rabo). Koyaya, wani lokacin ana iya samun siffofi daban-daban na kowane pixels - a wannan yanayin muna magana akai pixel rabo (PAR - daga Rabo Halayen Pixel).
Sanarwa ga uwar gida: DVD соответствует DAR 4 zu3
Kodayake ainihin ƙudurin DVD ɗin shine 704x480, har yanzu yana riƙe da 4: 3 yanayin rabo saboda PAR shine 10:11 (704x10 / 480x11).
Kuma a ƙarshe, zamu iya ƙayyade видео kamar jerin jerin n Frames na zamani lokaci, wanda za a iya la'akari da ƙarin girma. A n sannan shine adadin firam ko adadin firam a sakan daya (FPS - daga Frames a dakika).
Adadin ragowa a cikin daƙiƙa guda da ake buƙata don nuna bidiyo shine nasa gudun watsawa - bitrate.
bitrate = nisa * tsawo * zurfin bit * firam a sakan daya
Misali, 30fps, 24 bps, 480x240 bidiyo zai buƙaci 82,944,000 bps ko 82,944 Mbps (30x480x240x24) - amma wannan idan ba a yi amfani da hanyar matsawa ba.
Idan gudun canja wuri kusan akai, to ana kiranta m watsa gudun (Farashin CBR - daga m bit rate). Amma kuma yana iya bambanta, a cikin wannan yanayin ana kiran shi m darajar baud (VBR - daga m bit rate).
Wannan jadawali yana nuna ƙayyadaddun VBR, inda ba a ɓata rago da yawa a cikin yanayin firam mai duhu gaba ɗaya.
Da farko injiniyoyi sun ƙirƙiro wata hanya don ninka ƙimar firam ɗin nunin bidiyo ba tare da amfani da ƙarin bandwidth ba. Ana kiran wannan hanyar da video interlaced; Ainihin, yana aika rabin allon a cikin "frame" na farko da sauran rabi a cikin "frame" na gaba.
A halin yanzu, an fi yin amfani da al'amuran fasahohin bincike na ci gaba. Hanya ce ta nunawa, adanawa ko watsa hotuna masu motsi wanda aka zana duk layin kowane firam a jere.
To! Yanzu muna sane da yadda ake wakilta hoto ta hanyar lambobi, yadda aka tsara launukansa, ragi nawa muke kashewa a cikin dakika ɗaya don nuna bidiyo, idan ƙimar bit ɗin ya kasance akai-akai (CBR) ko mai canzawa (VBR). Mun san game da ƙudurin da aka bayar ta amfani da ƙimar firam ɗin da aka bayar, mun saba da wasu sharuɗɗa da yawa, kamar bidiyo mai haɗaka, PAR da wasu wasu.
Cire redundancy
An sani cewa bidiyo ba tare da matsawa ba za a iya amfani da shi kullum. Bidiyo na tsawon sa'a guda a ƙudurin 720p da firam 30 a cikin daƙiƙa guda zai ɗauki 278 GB. Mun isa wannan ƙimar ta ninka 1280 x 720 x 24 x 30 x 3600 (nisa, tsawo, ragowa kowane pixel, FPS da lokaci a cikin daƙiƙa).
Amfani algorithms matsawa marasa asara, kamar DEFLATE (amfani da PKZIP, Gzip da PNG), ba zai rage yawan bandwidth da ake buƙata ba. Dole ne mu nemi wasu hanyoyin da za a damfara bidiyo.
Don yin wannan, zaku iya amfani da fasalin hangen nesanmu. Mun fi kyau a bambanta haske fiye da launi. Bidiyo jerin jerin hotuna ne masu maimaita kan lokaci. Akwai ƙananan bambance-bambance tsakanin firam ɗin kusa na wuri ɗaya. Bugu da ƙari, kowane firam ɗin ya ƙunshi wurare da yawa ta amfani da launi ɗaya (ko kama).
Launi, haske da idanunmu
Idanuwanmu sun fi kula da haske fiye da launi. Kuna iya ganin wannan da kanku ta hanyar kallon wannan hoton.
Idan ba ku ga cewa a gefen hagu rabin hoton launukan murabba'i A и B a zahiri iri ɗaya ne, to wannan al'ada ce. Kwakwalwarmu tana tilasta mana mu mai da hankali ga haske da inuwa maimakon launi. A gefen dama tsakanin murabba'ai da aka keɓe akwai mai tsalle mai launi iri ɗaya - don haka mu (watau kwakwalwarmu) cikin sauƙin ƙayyade cewa, a zahiri, launi ɗaya ne.
Mu duba (a saukake) yadda idanunmu ke aiki. Ido wani hadadden gabo ne da ya kunshi sassa da dama. Koyaya, mun fi sha'awar mazugi da sanduna. Idon yana dauke da sanduna miliyan 120 da mazugi miliyan 6.
Bari mu yi la'akari da fahimtar launi da haske a matsayin ayyuka daban-daban na wasu sassan ido (a gaskiya, duk abin da ya fi rikitarwa, amma za mu sauƙaƙa shi). Kwayoyin sanduna galibi suna da alhakin haske, yayin da ƙwayoyin mazugi ke da alhakin launi. An raba mazugi zuwa nau'i uku, dangane da launin launi da suka ƙunshi: S-cones (blue), M-cones (kore), da L-cones (ja).
Tun da muna da sanduna da yawa (haske) fiye da mazugi (launi), za mu iya kammala cewa mun fi iya bambance sauye-sauye tsakanin duhu da haske fiye da launuka.
Dabarun Hankali na Bambanci
Masu bincike a cikin ilimin halin dan adam na gwaji da sauran fannoni da yawa sun haɓaka ka'idoji da yawa na hangen nesa na ɗan adam. Kuma ana kiran daya daga cikinsu bambanci ayyuka na hankali. Suna da alaƙa da haske na sarari da na ɗan lokaci. A takaice dai, game da canje-canje nawa ake buƙata kafin mai duba ya lura dasu. Kula da jam'in kalmar "aiki". Wannan shi ne saboda gaskiyar cewa za mu iya auna bambancin ayyuka na hankali ba kawai don hotuna baƙar fata da fari ba, har ma ga masu launi. Sakamakon waɗannan gwaje-gwajen ya nuna cewa a mafi yawan lokuta idanuwanmu sun fi kula da haske fiye da launi.
Tun da mun san cewa mun fi kula da hasken hoto, za mu iya ƙoƙarin yin amfani da wannan gaskiyar.
Samfurin launi
Mun gano kadan yadda ake aiki tare da hotunan launi ta amfani da tsarin RGB. Akwai kuma wasu model. Akwai samfurin da ke raba haske da chroma kuma an san shi da YCbCr. Af, akwai wasu samfurori da ke yin irin wannan rabo, amma za mu yi la'akari da wannan kawai.
A cikin wannan samfurin launi Y wakilci ne na haske, kuma yana amfani da tashoshi masu launi guda biyu: Cb (mai albarka blue) da Cr (mai arziki ja). Ana iya samun YCbCr daga RGB, kuma juyawa baya yana yiwuwa. Yin amfani da wannan ƙirar za mu iya ƙirƙirar cikakkun hotuna masu launi kamar yadda muke gani a ƙasa:
Canza tsakanin YCbCr da RGB
Wani zai ƙi: ta yaya zai yiwu a sami duk launuka idan ba a yi amfani da kore ba?
Don amsa wannan tambayar, bari mu canza RGB zuwa YCbCr. Bari mu yi amfani da ƙididdiga da aka karɓa a cikin ma'auni BT. 601, wanda naúrar ta ba da shawarar ITU-R. Wannan rarrabuwa tana saita ƙa'idodi don bidiyo na dijital. Misali: menene 4K? Menene yakamata ya zama ƙimar firam, ƙuduri, ƙirar launi?
Da farko bari mu lissafta haske. Bari mu yi amfani da madaidaicin da ITU ya gabatar kuma mu maye gurbin ƙimar RGB.
Y = 0.299R + 0.587G + 0.114B
Bayan muna da haske, za mu raba launin shuɗi da ja:
Cb = 0.564B - Y)
Cr = 0.713R - Y)
Kuma za mu iya juyawa baya har ma da samun kore ta amfani da YCbCr:
R = Y + 1.402Cr
B = Y + 1.772Cb
G = Y - 0.344Cb - 0.714Cr
Yawanci, nuni (masu duba, TV, fuska, da sauransu) suna amfani da ƙirar RGB kawai. Amma wannan model za a iya shirya ta hanyoyi daban-daban:
Samfuran launi
Tare da hoton da aka wakilta azaman haɗin haske da chrominance, za mu iya amfani da tsarin gani na ɗan adam mafi girman hankali ga haske fiye da chrominance ta zaɓin cire bayanai. Samfuran chroma hanya ce ta ɓoye hotuna ta amfani da ƙarancin ƙuduri don chroma fiye da haske.
Nawa ne ya halatta a rage ƙudurin launi?! Ya bayyana cewa akwai wasu zane-zane waɗanda ke bayyana yadda ake ɗaukar ƙuduri da haɗawa (Sakamakon Launi = Y + Cb + Cr).
Wadannan tsare-tsare an san su downsampling tsarin kuma an bayyana su azaman rabo mai ninki 3 - a:x:y, wanda ke ƙayyade adadin samfurori na haske da siginar bambancin launi.
a - mizanin samfurin kwance (yawanci daidai yake da 4)
x - adadin samfuran chroma a jere na farko na pixels (ƙudurin kwance dangane da a)
y - adadin canje-canje a samfuran chroma tsakanin layuka na farko da na biyu na pixels.
Banda shi ne 4:1:0, samar da samfurin chroma guda ɗaya a cikin kowane 4-by-4 luminance ƙuduri block.
Shirye-shiryen gama-gari da ake amfani da su a cikin codecs na zamani:
- 4:4:4 (babu downsampling)
- 4:2:2
- 4:1:1
- 4:2:0
- 4:1:0
- 3:1:1
YCbCr 4: 2: 0 - misalin haɗin gwiwa
Anan ga hoton da aka haɗa ta amfani da YCbCr 4:2:0. Lura cewa muna kashe rago 12 ne kawai akan kowane pixel.
Wannan shine yadda hoton daya yake kama da shi, wanda aka lullube shi tare da manyan nau'ikan samfurin launi. Layi na farko shine YCbCr na ƙarshe, layin ƙasa yana nuna ƙudurin chroma. Kyakkyawan sakamako mai kyau, la'akari da ƙarancin asarar inganci.
Ka tuna lokacin da muka ƙidaya 278 GB na sararin ajiya don adana fayil ɗin bidiyo na tsawon sa'a guda a ƙudurin 720p da firam 30 a sakan daya? Idan muka yi amfani da YCbCr 4: 2: 0, to wannan girman za a rage shi da rabi - 139 GB. Ya zuwa yanzu, har yanzu yana da nisa daga sakamako mai karɓuwa.
Kuna iya samun tarihin YCbCr da kanku ta amfani da FFmpeg. A cikin wannan hoton, shuɗi ya mamaye ja, wanda ke bayyane a fili a cikin histogram kanta.
Launi, haske, gamut launi - bita na bidiyo
Muna ba da shawarar kallon wannan bidiyo mai ban mamaki. Yana bayyana menene haske, kuma gabaɗaya duk ɗigon dige-dige ne ё game da haske da launi.
Nau'in Tsari
Mu ci gaba. Bari mu yi ƙoƙari mu kawar da raguwar lokaci. Amma da farko, bari mu ayyana wasu kalmomi na asali. Bari mu ce muna da fim mai firam 30 a cikin daƙiƙa guda, ga firam ɗinsa na farko guda 4:
Za mu iya ganin maimaituwa da yawa a cikin firam ɗin: misali, bangon shuɗi wanda baya canzawa daga firam zuwa firam. Don magance wannan matsalar, za mu iya rarraba su a hankali zuwa nau'ikan firam guda uku.
I-fram (Intro Frame)
I-frame (firam na nuni, firam ɗin maɓalli, firam na ciki) yana ƙunshe da kansa. Ko da kuwa abin da kuke son gani, I-frame ainihin hoto ne a tsaye. Firam na farko yawanci I-frame ne, amma za mu kiyaye I-frames akai-akai har ma a tsakanin firam ɗin farko.
P-fram (PFrame da aka gyara)
P-frame (firam ɗin tsinkaya) yana amfani da gaskiyar cewa kusan koyaushe ana iya sake yin hoton na yanzu ta amfani da firam ɗin da ya gabata. Misali, a cikin firam na biyu canjin kawai shine ƙwallon yana tafiya gaba. Za mu iya samun firam 2 ta hanyar ɗan gyaggyara firam 1, kawai ta amfani da bambanci tsakanin waɗannan firam ɗin. Don gina frame 2, muna komawa zuwa firam ɗin da ya gabata 1.
←
B-frame (BI-predictive Frame)
Me game da hanyoyin haɗin kai ba kawai na baya ba, har ma zuwa firam ɗin gaba don samar da matsi mafi kyau?! Wannan shine ainihin tsarin B-frame (firam ɗin bidirectional).
← →
Matsakaicin fitarwa
Ana amfani da waɗannan nau'ikan firam don samar da mafi kyawun yuwuwar matsawa. Za mu duba yadda hakan ke faruwa a sashe na gaba. A yanzu, bari mu lura cewa mafi "tsada" dangane da ƙwaƙwalwar ajiya da ake cinyewa shine I-frame, P-frame yana da rahusa sosai, amma mafi kyawun zaɓi don bidiyo shine B-frame.
Ragewar ɗan lokaci (hasashen tsaka-tsaki)
Bari mu kalli irin zaɓuɓɓukan da muke da su don rage maimaitawa cikin lokaci. Za mu iya magance wannan nau'in sakewa ta amfani da hanyoyin tsinkaya.
Za mu yi ƙoƙarin kashe kaɗan kaɗan gwargwadon yuwuwa don ɓoye jerin firam ɗin 0 da 1.
Za mu iya samarwa raguwa, kawai muna cire firam 1 daga firam 0. Muna samun firam 1, kawai muna amfani da bambanci tsakaninsa da firam ɗin da ya gabata, a zahiri muna ɓoye ragowar sakamakon.
Amma idan na gaya muku cewa akwai wata hanya mafi kyau wacce ke amfani da ƴan kaɗan kaɗan?! Da farko, bari mu karya firam 0 cikin madaidaicin grid wanda ya ƙunshi tubalan. Sannan za mu yi ƙoƙarin daidaita tubalan daga firam 0 tare da firam 1. A wasu kalmomi, za mu ƙididdige motsi tsakanin firam ɗin.
Daga Wikipedia - toshe motsi ramuwa
Toshe ramuwa na motsi yana raba firam ɗin na yanzu zuwa ɓangarorin da ba su cika ba kuma vector ramuwa na motsi ya ba da rahoton asalin tubalan (rashin fahimta na gama gari shine cewa. da suka gabata An raba firam ɗin zuwa ɓangarorin da ba su haɗa juna ba, kuma masu raɗaɗin motsi suna faɗin inda waɗannan tubalan ke tafiya. Amma a gaskiya ma, ita ce sauran hanyar - ba tsarin da ya gabata ba ne aka yi nazari, amma na gaba; ba a bayyana inda tubalan ke motsawa ba, amma inda suka fito). Yawanci tushen toshewar suna haɗuwa a cikin firam ɗin tushen. Wasu algorithms na matsa bidiyo suna haɗa firam ɗin na yanzu daga sassan ba ko ɗaya ba, amma firam ɗin da aka watsa a baya.
A lokacin aikin tantancewa, mun ga cewa ƙwallon ya tashi daga (x= 0, y=25) ku (x= 6, y=26), daraja x и y ƙayyade motsin motsi. Wani mataki da za mu iya yi don adana ragowa shine don ɓoye kawai bambancin motsin motsi tsakanin matsayi na ƙarshe da wanda aka annabta, don haka motsin motsi na ƙarshe zai kasance (x=6-0=6, y=26-25=1). ).
A cikin yanayi na ainihi, wannan ƙwallon za a raba shi n toshe, amma wannan baya canza ainihin lamarin.
Abubuwan da ke cikin firam ɗin suna motsawa cikin girma uku, don haka lokacin da ƙwallon ya motsa, zai iya zama ƙarami na gani (ko girma idan ya matsa zuwa mai kallo). Yana da al'ada cewa ba za a sami cikakkiyar daidaito tsakanin tubalan ba. Anan ga hadadden ra'ayi game da kiyasin mu da ainihin hoto.
Amma muna ganin cewa lokacin da muke amfani da kimanta motsi, akwai ƙarancin bayanai don yin coding fiye da lokacin amfani da hanya mafi sauƙi na ƙididdige delta tsakanin firam ɗin.
Yaya ainihin ramuwa na motsi zai yi kama
Ana amfani da wannan dabarar zuwa duk tubalan lokaci guda. Sau da yawa ƙwallon ƙwallon mu mai motsi na sharadi za a raba shi zuwa tubalan da yawa lokaci ɗaya.
Kuna iya jin daɗin waɗannan ra'ayoyin da kanku ta amfani da su .
Don ganin motsin motsi, zaku iya ƙirƙirar bidiyon hasashen waje ta amfani da .
Hakanan zaka iya amfani (an biya, amma akwai gwaji kyauta wanda ke iyakance ga firam goma na farko kawai).
Rage sarari (hasashen ciki)
Idan muka bincika kowane firam a cikin bidiyo, za mu sami wurare masu alaƙa da yawa.
Bari mu bi ta wannan misalin. Wannan yanayin ya ƙunshi launuka masu launin shuɗi da fari.
Wannan I-frame ne. Ba za mu iya ɗaukar firam ɗin baya don tsinkaya ba, amma za mu iya damfara shi. Bari mu shigar da zaɓi na jan block. Idan muka dubi makwabta, za mu lura cewa akwai wasu yanayi na launi a kusa da shi.
Muna ɗauka cewa launuka suna bazuwa a tsaye a cikin firam. Wanda ke nufin launin pixels da ba a san su ba zai ƙunshi ƙimar maƙwabtansa.
Irin wannan hasashen na iya zama ba daidai ba. A saboda wannan dalili ne kuke buƙatar amfani da wannan hanyar (hasashen ciki), sannan ku cire ainihin ƙimar. Wannan zai ba mu ragowar toshe, wanda zai haifar da matrix da yawa fiye da na asali.
Idan kuna son yin aiki tare da tsinkayar ciki, zaku iya ƙirƙirar bidiyo na macroblocks da tsinkayar su ta amfani da ffmpeg. Don fahimtar ma'anar kowane launi toshe, dole ne ku karanta takaddun ffmpeg.
Ko kuna iya amfani da Intel Video Pro Analyzer (kamar yadda na ambata a sama, sigar gwaji ta kyauta tana iyakance ga firam ɗin 10 na farko, amma wannan zai ishe ku da farko).
Kashi na biyu:
source: www.habr.com