Automa'c Radiometric Calibra'on from Mo'on Images Ricardo R. - - PowerPoint PPT Presentation

automa c radiometric calibra on
SMART_READER_LITE
LIVE PREVIEW

Automa'c Radiometric Calibra'on from Mo'on Images Ricardo R. - - PowerPoint PPT Presentation

Automa'c Radiometric Calibra'on from Mo'on Images Ricardo R. Figueroa Assistant Professor, Mo'on Picture Science GCCIS Part-'me PhD Student Jinwei Gu,


slide-1
SLIDE 1

Automa'c ¡Radiometric ¡Calibra'on ¡ ¡from ¡Mo'on ¡Images ¡

Ricardo ¡R. ¡Figueroa ¡

Assistant ¡Professor, ¡Mo'on ¡Picture ¡Science ¡ GCCIS ¡Part-­‑'me ¡PhD ¡Student ¡

Jinwei ¡Gu, ¡Pengchen ¡Shi ¡

Advisors ¡

slide-2
SLIDE 2

Topics ¡ ¡

  • Mo'va'on ¡
  • Image ¡Interchange ¡Framework ¡(IIF) ¡

Background ¡

  • State ¡of ¡the ¡Art ¡on ¡Response ¡Func'on ¡

Es'ma'on ¡Methods ¡

  • Methodology ¡
  • Ini'al ¡Results ¡
  • Path ¡Forward ¡

2 ¡

slide-3
SLIDE 3

Mo'va'on ¡

  • Understanding ¡of ¡the ¡camera ¡response ¡

func'on ¡provides: ¡ ¡

– Camera ¡characteriza'on ¡ – Irradiance/Radiance ¡from ¡measurements ¡ – Different ¡source ¡matching ¡(IIF) ¡ – Surpass ¡current ¡electronic ¡camera ¡limita'ons ¡ – HDR ¡imaging ¡ ¡

3 ¡

slide-4
SLIDE 4

Mo'va'on ¡-­‑ ¡Why? ¡

  • Example ¡-­‑ ¡Electronic ¡camera ¡limita'ons ¡

Also ¡known ¡as ¡ Gamma ¡correc'on ¡ curves ¡

4 ¡ Graph ¡borrowed ¡from ¡“Best ¡Prac'ces ¡Guide ¡to ¡Digital ¡Cinematography ¡Using ¡Panasonic ¡Professional ¡HD ¡Cameras” ¡ ¡ ¡

slide-5
SLIDE 5

Mo'va'on ¡-­‑ ¡IIF ¡

5 ¡

slide-6
SLIDE 6

Mo'va'on ¡-­‑ ¡IIF ¡

6 ¡

slide-7
SLIDE 7

Mo'va'on ¡– ¡Research ¡Problem ¡

  • From ¡Radiance ¡to ¡Code ¡Values ¡

7 ¡

slide-8
SLIDE 8

Background ¡-­‑ ¡Previous ¡Work ¡

  • Radiometric ¡Calibra'on ¡Methods ¡

– Direct ¡chart ¡measurement ¡ – Mul'ple ¡exposures ¡ – Single ¡image ¡(physical ¡image ¡a_ribute) ¡ – Photo ¡collec'ons ¡(aggregate ¡sta's'c) ¡

8 ¡

slide-9
SLIDE 9

Background ¡-­‑ ¡Response ¡Func'on ¡ Es'ma'on ¡

  • Direct ¡measurement ¡

Opto-­‑Electronic ¡Camera ¡Func'on ¡CHART ¡

9 ¡

3 ¡ 2 ¡ 1 ¡ 1 ¡ 1 ¡ 1 ¡ 2 ¡ 2 ¡ 2 ¡ 3 ¡ 3 ¡ 3 ¡

slide-10
SLIDE 10

Background ¡-­‑ ¡Response ¡Func'on ¡ Es'ma'on ¡

  • Mul'ple ¡exposures ¡

10 ¡

slide-11
SLIDE 11
  • Automa'c ¡Camera ¡Func'on ¡Calibra'on ¡

– Debevec ¡and ¡Malik ¡[2] ¡ – Mitsunaga ¡and ¡Nayar ¡[3] ¡

¡

Background ¡-­‑ ¡Response ¡Func'on ¡ Es'ma'on ¡

M = f I

( )

g = f −1 g mA

( ) = kg mB ( )

Intensity ¡ Image ¡Irradiance ¡ Camera ¡Response ¡Func'on ¡ Exposure ¡Ra'o ¡ Measured ¡Intensi'es ¡in ¡corresponding ¡images ¡ Inverse ¡Camera ¡Response ¡Func'on ¡ Exponen'al ¡ambiguity ¡[6] ¡ Rough ¡ini'al ¡es'mate ¡ Assump'ons ¡about ¡model ¡

11 ¡

slide-12
SLIDE 12
  • From ¡Radiance ¡to ¡Code ¡Values ¡and ¡Back ¡

– Popular ¡camera ¡response ¡model ¡by ¡Grossberg ¡and ¡ Nayar ¡[6] ¡

Background ¡-­‑ ¡Response ¡Func'on ¡ Es'ma'on ¡

f I

( ) = h0 I ( )+

wnhn I

( )

n=1 N

Image ¡Irradiance ¡ Average ¡Camera ¡Response ¡Func'on ¡ Principal ¡Components ¡ Unknown ¡Coefficients ¡

12 ¡

slide-13
SLIDE 13

Background ¡-­‑ ¡Response ¡Func'on ¡ Es'ma'on ¡

  • Single ¡Image ¡Analysis ¡[4] ¡ ¡

– Measured ¡edge ¡color ¡distribu'ons ¡ – Minimiza'on ¡of ¡distance ¡from ¡f-­‑1(M) ¡to ¡I ¡

13 ¡

slide-14
SLIDE 14

Methodology ¡

  • Expanded ¡Lin ¡et ¡al. ¡[4] ¡work ¡to ¡include ¡mul'ple ¡

frames ¡

– How? ¡

  • Look ¡at ¡color ¡edges ¡across ¡mul'ple ¡frames ¡
  • Color ¡edge ¡detec'on ¡
  • Add ¡correla'on ¡constraint ¡to ¡response ¡es'ma'on ¡
  • Do ¡we ¡need ¡a ¡prior ¡in ¡the ¡es'ma'on ¡

– Why? ¡

  • Increase ¡amount ¡of ¡color ¡edge ¡data ¡available ¡
  • Avoid ¡binary ¡response ¡ ¡
  • Addi'onal ¡data ¡might ¡be ¡sufficient ¡for ¡es'ma'on ¡

¡

14 ¡

slide-15
SLIDE 15

Methodology ¡

  • Distance ¡calcula'on ¡

¡

  • Bayesian ¡Es'ma'on ¡

15 ¡

D g;Ω

( ) =

| g M1

( )− g M2 ( )

# $ % &× g M1

( )− g M p

( )

# $ % &| | g M1

( )− g M2 ( ) |

Ω

g = argmin λD g;Ω

( )− log p g ( )

Prior ¡– ¡GMM ¡of ¡DoRF ¡[6] ¡

slide-16
SLIDE 16

Ini'al ¡Steps ¡-­‑ ¡Camera/Sensor ¡Data ¡ Collected ¡

  • Panasonic ¡GH2 ¡
  • Canon ¡5D ¡MarkII ¡
  • Arri ¡D-­‑21 ¡(2 ¡modes) ¡
  • Eastman ¡Kodak ¡7213 ¡(200T) ¡

16 ¡

slide-17
SLIDE 17

Ini'al ¡Steps ¡-­‑ ¡Camera/Sensor ¡Data ¡ Collected ¡

17 ¡

slide-18
SLIDE 18

Ini'al ¡Results ¡

18 ¡

Camera ¡Response ¡Es'mates ¡ Camera ¡Response ¡Es'mates ¡w/o ¡Prior ¡

slide-19
SLIDE 19

Path ¡Forward ¡

  • Use ¡video ¡footage ¡of ¡addi'onal ¡camera ¡

models ¡

– 24 ¡or ¡30 ¡fps ¡provided ¡

  • Study ¡demosaicing ¡algorithms ¡on ¡sets ¡selected ¡
  • Understand ¡image ¡sequence ¡sta's'cs ¡
  • btained ¡from ¡this ¡footage ¡

– Obtain ¡different ¡priors ¡for ¡op'miza'on ¡

  • Automa'cally ¡obtain ¡accurate ¡camera ¡

response ¡func'on ¡from ¡mo'on ¡images ¡

19 ¡

slide-20
SLIDE 20

References ¡

  • [1] ¡h_p://www.oscars.org/science-­‑technology/council/projects/iif.html ¡
  • [2] ¡Paul ¡E. ¡Debevec ¡and ¡Jitendra ¡Malik. ¡Recovering ¡high ¡dynamic ¡range ¡radiance ¡maps ¡from ¡
  • photographs. ¡In ¡Proceedings ¡of ¡the ¡24th ¡annual ¡conference ¡on ¡Computer ¡graphics ¡and ¡

interac've ¡techniques, ¡SIGGRAPH ¡’97, ¡pages ¡369–378, ¡New ¡York, ¡NY, ¡USA, ¡1997. ¡

  • [3] ¡Mitsunaga, ¡T.; ¡Nayar, ¡S.K.; ¡ ¡"Radiometric ¡self ¡calibra'on," ¡Computer ¡Vision ¡and ¡Pa_ern ¡

Recogni'on, ¡1999. ¡IEEE ¡Computer ¡Society ¡Conference ¡on. ¡, ¡vol.1, ¡no., ¡pp.2 ¡vol. ¡(xxiii +637+663), ¡1999 ¡

  • [4] ¡Lin, ¡S.; ¡Jinwei ¡Gu; ¡Yamazaki, ¡S.; ¡Heung-­‑Yeung ¡Shum; ¡, ¡"Radiometric ¡calibra'on ¡from ¡a ¡

single ¡image," ¡Computer ¡Vision ¡and ¡Pa_ern ¡Recogni'on, ¡2004. ¡CVPR ¡2004. ¡Proceedings ¡of ¡ the ¡2004 ¡IEEE ¡Computer ¡Society ¡Conference ¡on ¡, ¡vol.2, ¡no., ¡pp. ¡II-­‑938-­‑ ¡II-­‑945 ¡Vol.2, ¡27 ¡June-­‑2 ¡ July ¡2004 ¡

  • [5] ¡Sujit ¡Kuthirummal, ¡Aseem ¡Agarwala, ¡Dan ¡B ¡Goldman, ¡and ¡Shree ¡K. ¡Nayar. ¡“Priors ¡for ¡large ¡

photo ¡collec'ons ¡and ¡what ¡they ¡reveal ¡about ¡Cameras”, ¡2008. ¡

  • [6] ¡Grossberg, ¡M.D.; ¡Nayar, ¡S.K.; ¡ ¡"Determining ¡the ¡camera ¡response ¡from ¡images: ¡what ¡is ¡

knowable?" ¡Pa_ern ¡Analysis ¡and ¡Machine ¡Intelligence, ¡IEEE ¡Transac'ons ¡on ¡, ¡vol.25, ¡no.11, ¡

  • pp. ¡1455-­‑ ¡1467, ¡Nov. ¡2003 ¡

20 ¡