SlideShare ist ein Scribd-Unternehmen logo
1 von 35
Downloaden Sie, um offline zu lesen
Understanding	
  and	
  Improving	
  
the	
  Realism	
  of	
  Image	
  Composites	
  

               Aliya	
  Ibragimova	
  
                           	
  
             University	
  of	
  Fribourg	
  
Agenda	
  	
  
•    Realis7c	
  Image	
  composi7ng	
  
•    Iden7fying	
  key	
  sta7s7cs	
  	
  
•    Human	
  subject	
  experiments	
  
•    Algorithm	
  	
  
•    Results	
  
Image	
  composi7ng	
  
Composi7ng	
  procedure	
  
1.	
  	
  




               Foreground	
            Alpha	
  maAe	
  

2.	
  




             New	
  background	
  

                                         Composite	
  
Color	
  Transfer	
  Technique	
  (CTT)	
  	
  
                         Reinhard	
  et	
  al.	
  2001	
  




	
  ‘Match	
  color’	
  feature	
  of	
  Photoshop	
  
	
  
Color	
  transfer	
  technique:	
  limita7ons	
  




  Cut-­‐and-­‐paste	
                   Match	
  Color	
  

Conflates	
  the	
  effects	
  of	
  reflectance	
  and	
  illumina7on	
  
Improvements	
  of	
  CTT	
  (ColorComp)	
  	
  
              Lalonde	
  and	
  Effros	
  2007	
  

•  Analyze	
  huge	
  dataset	
  of	
  natural	
  images	
  :	
  
     difference	
  in	
  distribu7on	
  of	
  realis7c	
  and	
  
     unrealis7c	
  images	
  
•  Recolor	
  regions	
  for	
  realis7c	
  composi7ng	
  
	
  
Limita7ons:	
  requires	
  a	
  large	
  dataset,	
  depends	
  
on	
  the	
  presence	
  of	
  images	
  that	
  are	
  similar	
  to	
  the	
  
target	
  	
  
#	
  of	
  pixels	
          Professional	
  compositors	
  




                              Shadows	
      Midtones	
          Highlights	
  
                        Brightness	
  

                 •  Isolate	
  highlights	
  and	
  match	
  their	
  colors	
  and	
  
                    brightness	
  
                 •  Balance	
  mid-­‐tones	
  with	
  gamma	
  correc7on	
  
                 •  Match	
  the	
  shadow	
  regions	
  
Color	
  Harmony	
  
                     	
  Cohen-­‐Or	
  et	
  al.	
  2006	
  




Harmonic	
  colors	
  are	
  sets	
  of	
  colors	
  that	
  are	
  aesthe7cally	
  
pleasing	
  in	
  terms	
  of	
  human	
  visual	
  percep7on.	
  
Color	
  Harmony	
  
                    	
  Cohen-­‐Or	
  et	
  al.	
  2006	
  




Limita7ons:	
  obtained	
  images	
  are	
  not	
  necessary	
  realis7c,	
  
ignores	
  luminance	
  and	
  contrast,	
  the	
  approach	
  has	
  not	
  
been	
  quan7ta7vely	
  evaluated	
  
Alterna7ve	
  to	
  alpha	
  maAe:	
  seamlessly	
  blending	
  

•  Feathering	
  
•  Laplacian	
  pyramids	
  [Odgen	
  et	
  al.	
  1985]	
  
•  Gradient-­‐domain	
  composi7ng	
  [Perez	
  et	
  al.	
  
   2003]	
  
Alterna7ve	
  to	
  alpha	
  maAe:	
  seamlessly	
  blending	
  




         foreground	
                          background	
  




         Cut-­‐and-­‐paste	
                 Gradient-­‐domain	
  

  Limita7ons:	
  2	
  source	
  images	
  should	
  have	
  similar	
  
  colors	
  and	
  textures	
  
Problem	
  statement	
  
•  Which	
  sta7s7cs	
  control	
  realism?	
  
•  How	
  do	
  these	
  sta7s7cs	
  affect	
  human	
  
   judgment	
  of	
  realism?	
  
•  Automa7c	
  algorithm	
  to	
  improve	
  realism?	
  	
  
Good	
  sta7s7cs	
  
•  Highly	
  correlated	
  between	
  foreground	
  and	
  
   background	
  
•  Easy	
  to	
  adjust	
  
•  Independent	
  from	
  each	
  other	
  
Categories	
  of	
  sta7s7cal	
  measures	
  
•    Luminance	
  
•    Color	
  temperature	
  (CCT)	
  
•    Satura7on	
  
•    Local	
  contrast	
  
•    Hue	
  (circular	
  sta7s7cs)	
  	
  
Sta7s7cal	
  measures	
  



                                                                    Standard	
  devia7on	
  
#	
  of	
  pixels	
  




                                                         mean	
  
                             Brightness	
  
                        •     Mean	
  
                        •     Standard	
  devia7on	
  
                        •     High	
  
                        •     Low	
  
                        •     Kurtosis	
  
                        •     Entropy	
  
Find	
  correla7on	
  
•  Pearson	
  correla7on	
  coefficient	
  
•  Standard	
  devia7on	
  of	
  offset	
  δi	
  =	
  Mif	
  –	
  Mib	
  
	
  	
  
                                             M	
  –	
  measure	
  
                                             f	
  –	
  foreground	
  
                                             b	
  –	
  background	
  
                                             i	
  –	
  sta7s7cs	
  
                                             	
  
Sta7s7cal	
  experiment	
  

•  Use	
  large	
  (4126	
  images)	
  labeled	
  dataset	
  
•  Select	
  the	
  most	
  correlated	
  sta7s7cs	
  
Luminance	
  
Color	
  Temperature	
  
Satura7on	
  
Sta7s7cal	
  experiment	
  
Results:	
  sta7s7cal	
  experiment	
  
	
  
•  	
  luminance,	
  color	
  temperature,	
  satura7on,	
  
     local	
  contrast	
  are	
  most	
  correlated	
  
•  	
  mean	
  of	
  zones	
  correlate	
  more	
  than	
  other	
  
     sta7s7cal	
  measures	
  
•  	
  mean	
  of	
  high	
  and	
  low	
  zones	
  correlate	
  more	
  
     than	
  mean	
  of	
  en7re	
  histogram	
  
Human	
  subjects	
  experiment	
  
Experiment	
  with	
  human	
  subjects	
  on	
  Amazon	
  
Mechanical	
  Turk	
  (MTurk)	
  	
  

•  20	
  natural	
  images	
  
•  3	
  key	
  sta7s7cs	
  (luminance,	
  color	
  temperature,	
  satura7on)	
  
Human	
  subject	
  experiments	
  



Foreground	
  luminance	
  




                              Background	
  luminance	
  
Human	
  subjects	
  experiment	
  
Human	
  subjects	
  experiment	
  
Automa7c	
  composite	
  adjustment	
  
•  Zone	
  selec7on	
  using	
  machine	
  learning	
  
     (random	
  forest	
  classifier)	
  
     	
   	
   	
  T	
  =	
  s	
  *	
  σg,	
  s	
  =	
  0.1	
  
     	
  Three	
  binary	
  classifiers:	
  
•  Pick	
  smallest	
  changes	
  if	
  mul7ple	
  zones	
  
•  Features	
  for	
  foreground	
  and	
  background	
  and	
  
     per	
  sta7s7c:	
  std,	
  skew,	
  kurt,	
  entropy,	
  p1,	
  p2	
  …
     p20.	
  	
  
	
  
Pipeline	
  
Input:	
  foreground	
  and	
  background	
  image	
  
1.  Match	
  H-­‐zone	
  of	
  contrast	
  using	
  S-­‐shape	
  
2.  Select	
  zone	
  and	
  adjust	
  mean	
  of	
  luminance	
  
3.  Select	
  zone	
  and	
  adjust	
  mean	
  of	
  CCT	
  
4.  Select	
  zone	
  and	
  adjust	
  satura7on	
  
	
  
Adjust	
  algorithm	
  is	
  greedy,	
  could	
  iterate	
  several	
  
7mes	
  if	
  needed	
  	
  
Results	
  and	
  evalua7on	
  




Cut-­‐and-­‐paste	
       Manual	
     MatchColor	
     ColorComp	
     Ours	
  
Results	
  and	
  evalua7on	
  




Cut-­‐and-­‐paste	
       Manual	
     MatchColor	
     ColorComp	
     Ours	
  
Results	
  and	
  evalua7on	
  
Results	
  and	
  evalua7on	
  
Thurstone’s	
  Law	
  of	
  Compara7ve	
  Judgement	
  
Results	
  and	
  evalua7on	
  
•  BeAer	
  than	
  previous	
  methods	
  
•  Close	
  to	
  manual	
  edi7ng	
  

Limita7ons:	
  
•  Religh7ng	
  
•  Theory	
  for	
  zone	
  selec7on	
  	
  
Ques7ons?	
  

Weitere ähnliche Inhalte

Was ist angesagt?

A Physical Approach to Moving Cast Shadow Detection (ICASSP 2009)
A Physical Approach to Moving Cast Shadow Detection (ICASSP 2009)A Physical Approach to Moving Cast Shadow Detection (ICASSP 2009)
A Physical Approach to Moving Cast Shadow Detection (ICASSP 2009)
Jia-Bin Huang
 
Estimating Human Pose from Occluded Images (ACCV 2009)
Estimating Human Pose from Occluded Images (ACCV 2009)Estimating Human Pose from Occluded Images (ACCV 2009)
Estimating Human Pose from Occluded Images (ACCV 2009)
Jia-Bin Huang
 
Tuto part2
Tuto part2Tuto part2
Tuto part2
Bo Li
 
Modern features-part-4-evaluation
Modern features-part-4-evaluationModern features-part-4-evaluation
Modern features-part-4-evaluation
zukun
 
igarss.pdf
igarss.pdfigarss.pdf
igarss.pdf
grssieee
 
Fingerprint _prem
Fingerprint _premFingerprint _prem
Fingerprint _prem
lgbl40
 
Unsupervised Change Detection in the Feature Space Using Kernels.pdf
Unsupervised Change Detection in the Feature Space Using Kernels.pdfUnsupervised Change Detection in the Feature Space Using Kernels.pdf
Unsupervised Change Detection in the Feature Space Using Kernels.pdf
grssieee
 

Was ist angesagt? (7)

A Physical Approach to Moving Cast Shadow Detection (ICASSP 2009)
A Physical Approach to Moving Cast Shadow Detection (ICASSP 2009)A Physical Approach to Moving Cast Shadow Detection (ICASSP 2009)
A Physical Approach to Moving Cast Shadow Detection (ICASSP 2009)
 
Estimating Human Pose from Occluded Images (ACCV 2009)
Estimating Human Pose from Occluded Images (ACCV 2009)Estimating Human Pose from Occluded Images (ACCV 2009)
Estimating Human Pose from Occluded Images (ACCV 2009)
 
Tuto part2
Tuto part2Tuto part2
Tuto part2
 
Modern features-part-4-evaluation
Modern features-part-4-evaluationModern features-part-4-evaluation
Modern features-part-4-evaluation
 
igarss.pdf
igarss.pdfigarss.pdf
igarss.pdf
 
Fingerprint _prem
Fingerprint _premFingerprint _prem
Fingerprint _prem
 
Unsupervised Change Detection in the Feature Space Using Kernels.pdf
Unsupervised Change Detection in the Feature Space Using Kernels.pdfUnsupervised Change Detection in the Feature Space Using Kernels.pdf
Unsupervised Change Detection in the Feature Space Using Kernels.pdf
 

Andere mochten auch (7)

Cat orgin ofmfr
Cat orgin ofmfrCat orgin ofmfr
Cat orgin ofmfr
 
Stop-the-world GCs on milticores
Stop-the-world GCs on milticoresStop-the-world GCs on milticores
Stop-the-world GCs on milticores
 
Starting a company in Qatar
Starting a company in QatarStarting a company in Qatar
Starting a company in Qatar
 
Sport tandem
Sport tandemSport tandem
Sport tandem
 
Guidelines toupload
Guidelines touploadGuidelines toupload
Guidelines toupload
 
manfaat dan kandungan alpukat
manfaat dan kandungan alpukatmanfaat dan kandungan alpukat
manfaat dan kandungan alpukat
 
Ah, lord god
Ah, lord godAh, lord god
Ah, lord god
 

Ähnlich wie Realism of image composits

presentation644v4
presentation644v4presentation644v4
presentation644v4
Maikon
 
IMAGE FUSION IN IMAGE PROCESSING
IMAGE FUSION IN IMAGE PROCESSINGIMAGE FUSION IN IMAGE PROCESSING
IMAGE FUSION IN IMAGE PROCESSING
garima0690
 
TOWARDS OPTIMALITY OF IMAGE SEGMENTATION PART- I
TOWARDS OPTIMALITY OF IMAGE SEGMENTATION PART- ITOWARDS OPTIMALITY OF IMAGE SEGMENTATION PART- I
TOWARDS OPTIMALITY OF IMAGE SEGMENTATION PART- I
Anish Acharya
 
Presentation at SMI 2023
Presentation at SMI 2023Presentation at SMI 2023
Presentation at SMI 2023
Joaquim Jorge
 
數位影像處理-期末報告
數位影像處理-期末報告數位影像處理-期末報告
數位影像處理-期末報告
Hesher Shih
 
Seminar_Thoracic EIT_UWO
Seminar_Thoracic EIT_UWOSeminar_Thoracic EIT_UWO
Seminar_Thoracic EIT_UWO
Peyman Rahmati
 
Surface Data Capture Image Matching
Surface Data Capture Image MatchingSurface Data Capture Image Matching
Surface Data Capture Image Matching
RASHID JAVED
 
Super resolution in deep learning era - Jaejun Yoo
Super resolution in deep learning era - Jaejun YooSuper resolution in deep learning era - Jaejun Yoo
Super resolution in deep learning era - Jaejun Yoo
JaeJun Yoo
 

Ähnlich wie Realism of image composits (20)

Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...
 
presentation644v4
presentation644v4presentation644v4
presentation644v4
 
CVPR2008 tutorial generalized pca
CVPR2008 tutorial generalized pcaCVPR2008 tutorial generalized pca
CVPR2008 tutorial generalized pca
 
IMAGE FUSION IN IMAGE PROCESSING
IMAGE FUSION IN IMAGE PROCESSINGIMAGE FUSION IN IMAGE PROCESSING
IMAGE FUSION IN IMAGE PROCESSING
 
TOWARDS OPTIMALITY OF IMAGE SEGMENTATION PART- I
TOWARDS OPTIMALITY OF IMAGE SEGMENTATION PART- ITOWARDS OPTIMALITY OF IMAGE SEGMENTATION PART- I
TOWARDS OPTIMALITY OF IMAGE SEGMENTATION PART- I
 
Template matching03
Template matching03Template matching03
Template matching03
 
Global contrast based salient region detection
Global contrast based salient region detectionGlobal contrast based salient region detection
Global contrast based salient region detection
 
Presentation at SMI 2023
Presentation at SMI 2023Presentation at SMI 2023
Presentation at SMI 2023
 
Passive stereo vision with deep learning
Passive stereo vision with deep learningPassive stereo vision with deep learning
Passive stereo vision with deep learning
 
數位影像處理-期末報告
數位影像處理-期末報告數位影像處理-期末報告
數位影像處理-期末報告
 
06 image features
06 image features06 image features
06 image features
 
Digital Image Fundamentals
Digital Image FundamentalsDigital Image Fundamentals
Digital Image Fundamentals
 
Lec13 stereo converted
Lec13 stereo convertedLec13 stereo converted
Lec13 stereo converted
 
Searching Images with MPEG-7 (& MPEG-7 Like) Powered Localized dEscriptors (S...
Searching Images with MPEG-7 (& MPEG-7 Like) Powered Localized dEscriptors (S...Searching Images with MPEG-7 (& MPEG-7 Like) Powered Localized dEscriptors (S...
Searching Images with MPEG-7 (& MPEG-7 Like) Powered Localized dEscriptors (S...
 
Seminar_Thoracic EIT_UWO
Seminar_Thoracic EIT_UWOSeminar_Thoracic EIT_UWO
Seminar_Thoracic EIT_UWO
 
Surface Data Capture Image Matching
Surface Data Capture Image MatchingSurface Data Capture Image Matching
Surface Data Capture Image Matching
 
Analyzing color imaging failure on consumer-grade cameras
Analyzing color imaging failure on consumer-grade camerasAnalyzing color imaging failure on consumer-grade cameras
Analyzing color imaging failure on consumer-grade cameras
 
Methods of Manifold Learning for Dimension Reduction of Large Data Sets
Methods of Manifold Learning for Dimension Reduction of Large Data SetsMethods of Manifold Learning for Dimension Reduction of Large Data Sets
Methods of Manifold Learning for Dimension Reduction of Large Data Sets
 
Chap. 10 computational photography
Chap. 10 computational photographyChap. 10 computational photography
Chap. 10 computational photography
 
Super resolution in deep learning era - Jaejun Yoo
Super resolution in deep learning era - Jaejun YooSuper resolution in deep learning era - Jaejun Yoo
Super resolution in deep learning era - Jaejun Yoo
 

Kürzlich hochgeladen

+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
?#DUbAI#??##{{(☎️+971_581248768%)**%*]'#abortion pills for sale in dubai@
 
Artificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsArtificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and Myths
Joaquim Jorge
 

Kürzlich hochgeladen (20)

From Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time AutomationFrom Event to Action: Accelerate Your Decision Making with Real-Time Automation
From Event to Action: Accelerate Your Decision Making with Real-Time Automation
 
MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024
 
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost SavingRepurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
 
Data Cloud, More than a CDP by Matt Robison
Data Cloud, More than a CDP by Matt RobisonData Cloud, More than a CDP by Matt Robison
Data Cloud, More than a CDP by Matt Robison
 
The 7 Things I Know About Cyber Security After 25 Years | April 2024
The 7 Things I Know About Cyber Security After 25 Years | April 2024The 7 Things I Know About Cyber Security After 25 Years | April 2024
The 7 Things I Know About Cyber Security After 25 Years | April 2024
 
Top 5 Benefits OF Using Muvi Live Paywall For Live Streams
Top 5 Benefits OF Using Muvi Live Paywall For Live StreamsTop 5 Benefits OF Using Muvi Live Paywall For Live Streams
Top 5 Benefits OF Using Muvi Live Paywall For Live Streams
 
TrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
TrustArc Webinar - Stay Ahead of US State Data Privacy Law DevelopmentsTrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
TrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
 
Exploring the Future Potential of AI-Enabled Smartphone Processors
Exploring the Future Potential of AI-Enabled Smartphone ProcessorsExploring the Future Potential of AI-Enabled Smartphone Processors
Exploring the Future Potential of AI-Enabled Smartphone Processors
 
A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?
 
🐬 The future of MySQL is Postgres 🐘
🐬  The future of MySQL is Postgres   🐘🐬  The future of MySQL is Postgres   🐘
🐬 The future of MySQL is Postgres 🐘
 
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
 
Artificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and MythsArtificial Intelligence: Facts and Myths
Artificial Intelligence: Facts and Myths
 
Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)
 
Boost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivityBoost PC performance: How more available memory can improve productivity
Boost PC performance: How more available memory can improve productivity
 
Strategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherStrategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a Fresher
 
Automating Google Workspace (GWS) & more with Apps Script
Automating Google Workspace (GWS) & more with Apps ScriptAutomating Google Workspace (GWS) & more with Apps Script
Automating Google Workspace (GWS) & more with Apps Script
 
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
 
Apidays New York 2024 - The value of a flexible API Management solution for O...
Apidays New York 2024 - The value of a flexible API Management solution for O...Apidays New York 2024 - The value of a flexible API Management solution for O...
Apidays New York 2024 - The value of a flexible API Management solution for O...
 
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data DiscoveryTrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
 

Realism of image composits

  • 1. Understanding  and  Improving   the  Realism  of  Image  Composites   Aliya  Ibragimova     University  of  Fribourg  
  • 2. Agenda     •  Realis7c  Image  composi7ng   •  Iden7fying  key  sta7s7cs     •  Human  subject  experiments   •  Algorithm     •  Results  
  • 4. Composi7ng  procedure   1.     Foreground   Alpha  maAe   2.   New  background   Composite  
  • 5. Color  Transfer  Technique  (CTT)     Reinhard  et  al.  2001    ‘Match  color’  feature  of  Photoshop    
  • 6. Color  transfer  technique:  limita7ons   Cut-­‐and-­‐paste   Match  Color   Conflates  the  effects  of  reflectance  and  illumina7on  
  • 7. Improvements  of  CTT  (ColorComp)     Lalonde  and  Effros  2007   •  Analyze  huge  dataset  of  natural  images  :   difference  in  distribu7on  of  realis7c  and   unrealis7c  images   •  Recolor  regions  for  realis7c  composi7ng     Limita7ons:  requires  a  large  dataset,  depends   on  the  presence  of  images  that  are  similar  to  the   target    
  • 8. #  of  pixels   Professional  compositors   Shadows   Midtones   Highlights   Brightness   •  Isolate  highlights  and  match  their  colors  and   brightness   •  Balance  mid-­‐tones  with  gamma  correc7on   •  Match  the  shadow  regions  
  • 9. Color  Harmony    Cohen-­‐Or  et  al.  2006   Harmonic  colors  are  sets  of  colors  that  are  aesthe7cally   pleasing  in  terms  of  human  visual  percep7on.  
  • 10. Color  Harmony    Cohen-­‐Or  et  al.  2006   Limita7ons:  obtained  images  are  not  necessary  realis7c,   ignores  luminance  and  contrast,  the  approach  has  not   been  quan7ta7vely  evaluated  
  • 11. Alterna7ve  to  alpha  maAe:  seamlessly  blending   •  Feathering   •  Laplacian  pyramids  [Odgen  et  al.  1985]   •  Gradient-­‐domain  composi7ng  [Perez  et  al.   2003]  
  • 12. Alterna7ve  to  alpha  maAe:  seamlessly  blending   foreground   background   Cut-­‐and-­‐paste   Gradient-­‐domain   Limita7ons:  2  source  images  should  have  similar   colors  and  textures  
  • 13. Problem  statement   •  Which  sta7s7cs  control  realism?   •  How  do  these  sta7s7cs  affect  human   judgment  of  realism?   •  Automa7c  algorithm  to  improve  realism?    
  • 14. Good  sta7s7cs   •  Highly  correlated  between  foreground  and   background   •  Easy  to  adjust   •  Independent  from  each  other  
  • 15. Categories  of  sta7s7cal  measures   •  Luminance   •  Color  temperature  (CCT)   •  Satura7on   •  Local  contrast   •  Hue  (circular  sta7s7cs)    
  • 16. Sta7s7cal  measures   Standard  devia7on   #  of  pixels   mean   Brightness   •  Mean   •  Standard  devia7on   •  High   •  Low   •  Kurtosis   •  Entropy  
  • 17. Find  correla7on   •  Pearson  correla7on  coefficient   •  Standard  devia7on  of  offset  δi  =  Mif  –  Mib       M  –  measure   f  –  foreground   b  –  background   i  –  sta7s7cs    
  • 18. Sta7s7cal  experiment   •  Use  large  (4126  images)  labeled  dataset   •  Select  the  most  correlated  sta7s7cs  
  • 23. Results:  sta7s7cal  experiment     •   luminance,  color  temperature,  satura7on,   local  contrast  are  most  correlated   •   mean  of  zones  correlate  more  than  other   sta7s7cal  measures   •   mean  of  high  and  low  zones  correlate  more   than  mean  of  en7re  histogram  
  • 24. Human  subjects  experiment   Experiment  with  human  subjects  on  Amazon   Mechanical  Turk  (MTurk)     •  20  natural  images   •  3  key  sta7s7cs  (luminance,  color  temperature,  satura7on)  
  • 25. Human  subject  experiments   Foreground  luminance   Background  luminance  
  • 28. Automa7c  composite  adjustment   •  Zone  selec7on  using  machine  learning   (random  forest  classifier)        T  =  s  *  σg,  s  =  0.1    Three  binary  classifiers:   •  Pick  smallest  changes  if  mul7ple  zones   •  Features  for  foreground  and  background  and   per  sta7s7c:  std,  skew,  kurt,  entropy,  p1,  p2  … p20.      
  • 29. Pipeline   Input:  foreground  and  background  image   1.  Match  H-­‐zone  of  contrast  using  S-­‐shape   2.  Select  zone  and  adjust  mean  of  luminance   3.  Select  zone  and  adjust  mean  of  CCT   4.  Select  zone  and  adjust  satura7on     Adjust  algorithm  is  greedy,  could  iterate  several   7mes  if  needed    
  • 30. Results  and  evalua7on   Cut-­‐and-­‐paste   Manual   MatchColor   ColorComp   Ours  
  • 31. Results  and  evalua7on   Cut-­‐and-­‐paste   Manual   MatchColor   ColorComp   Ours  
  • 33. Results  and  evalua7on   Thurstone’s  Law  of  Compara7ve  Judgement  
  • 34. Results  and  evalua7on   •  BeAer  than  previous  methods   •  Close  to  manual  edi7ng   Limita7ons:   •  Religh7ng   •  Theory  for  zone  selec7on