Ce diaporama a bien été signalé.
Nous utilisons votre profil LinkedIn et vos données d’activité pour vous proposer des publicités personnalisées et pertinentes. Vous pouvez changer vos préférences de publicités à tout moment.
1/51
Introduction Our method Summary
Style-aware robust color transfer
H. Hristova O. Le Meur R. Cozot K. Bouatouch
IRISA ...
2/51
Introduction Our method Summary
Outline
1 Introduction
Context
Objective
Contributions
2 Our method
Algorithm
Evaluat...
3/51
Introduction Our method Summary
Context
Outline
1 Introduction
Context
Objective
Contributions
2 Our method
Algorithm...
4/51
Introduction Our method Summary
Context
Color transfer domain
Objective of color transfer
Color transfer aims at modi...
5/51
Introduction Our method Summary
Context
State-of-the-art global methods
 [Reinhard et al., 2001]
 Linear mapping
 L
Diagonal covariance matrix
 Gaussian distribution
 [Pitie et al., 2007]
 Linear mapping as a solution to Monge-Kantorovich...
6/51
Introduction Our method Summary
Context
State-of-the-art local methods
 [Tai et al., 2005]
 Linear mapping: [Reinhard...
3D image clustering
 Cluster mapping: maps the closest clusters in terms of their
luminance channel values
 Gaussian distr...
7/51
Introduction Our method Summary
Context
Limitations
Style-aware robust color transfer H. Hristova, O. Le Meur, R. Coz...
7/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
Style-aware robust color transfer H. Hristova...
7/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
InputimageReferenceimage[Reinhardetal.2001]
S...
8/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
Gaussian distribution
Style-aware robust colo...
8/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
Gaussian distribution
Style-aware robust colo...
9/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
Gaussian distribution
Clustering on the same
...
9/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
Gaussian distribution
Clustering on the same
...
10/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
Gaussian distribution
Clustering on the same...
10/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
Gaussian distribution
Clustering on the same...
11/51
Introduction Our method Summary
Context
Limitations
Style inconsistency
Gaussian distribution
Clustering on the same...
12/51
Introduction Our method Summary
Objective
Outline
1 Introduction
Context
Objective
Contributions
2 Our method
Algori...
13/51
Introduction Our method Summary
Objective
Our objective
Transferring the light and color features of a reference ima...
14/51
Introduction Our method Summary
Objective
Example
Input Reference
Style-aware robust color transfer H. Hristova, O. ...
15/51
Introduction Our method Summary
Objective
Example
Input Reference
[Pitieetal.,2007]
Style-aware robust color transfe...
16/51
Introduction Our method Summary
Objective
Example
Input Reference
[Pitieetal.,2007][Bonneeletal.,2013]
Style-aware r...
17/51
Introduction Our method Summary
Objective
Example
Input Reference
[Pitieetal.,2007][Bonneeletal.,2013]
Style-aware r...
18/51
Introduction Our method Summary
Contributions
Outline
1 Introduction
Context
Objective
Contributions
2 Our method
Al...
19/51
Introduction Our method Summary
Contributions
 Classi
cation of the input and reference images
 Light-based style images
 Colors-based style images
Style-aware robust color tra...
19/51
Introduction Our method Summary
Contributions
 Classi
cation of the input and reference images
 Light-based style images
 Colors-based style images
 Two ways of clustering:
 Lu...
19/51
Introduction Our method Summary
Contributions
 Classi
cation of the input and reference images
 Light-based style images
 Colors-based style images
 Two ways of clustering:
 Lu...
19/51
Introduction Our method Summary
Contributions
 Classi
cation of the input and reference images
 Light-based style images
 Colors-based style images
 Two ways of clustering:
 Lu...
20/51
Introduction Our method Summary
Algorithm
Outline
1 Introduction
Context
Objective
Contributions
2 Our method
Algori...
21/51
Introduction Our method Summary
Algorithm
Framework
light, colors
light, colors
Input
image
Reference
image
Classifi...
22/51
Introduction Our method Summary
Algorithm
Automatic image classi
cation system
Colors-based style images
Images which color information is sucient enough to well-de
ne
at least two dierent and signi
cant colors.
Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
23/51
Introduction Our method Summary
Algorithm
Automatic image classi
cation system
Light-based style images
Images which are not classi
ed as colors-based style images are
classi
ed as light-based style images. Their light features are more
meaningful than their color features.
Style-aware robust col...
24/51
Introduction Our method Summary
Algorithm
Framework
light, colors
light, colors
Input
image
Reference
image
Classifi...
25/51
Introduction Our method Summary
Algorithm
Clustering: Gaussian mixture models
Light-based style image Colors-based s...
26/51
Introduction Our method Summary
Algorithm
Framework
light, colors
light, colors
Input
image
Reference
image
Classifi...
27/51
Introduction Our method Summary
Algorithm
Pseudo code algorithm
for k = 1; : : : ; N do
if Light to Colors then
ILab...
28/51
Introduction Our method Summary
Algorithm
Mapping policies
for k = 1; : : : ; N do
if Light to Colors then
ILab := F...
29/51
Introduction Our method Summary
Algorithm
Mapping policies
for k = 1; : : : ; N do
if Light to Colors then
ILab := F...
30/51
Introduction Our method Summary
Algorithm
Mapping policies
for k = 1; : : : ; N do
if Light to Colors then
ILab := F...
31/51
Introduction Our method Summary
Algorithm
Mapping policies
for k = 1; : : : ; N do
if Light to Colors then
ILab := F...
32/51
Introduction Our method Summary
Algorithm
Color transformation
for k = 1; : : : ; N do
if Light to Colors then
ILab ...
33/51
Introduction Our method Summary
Algorithm
Color transformation
for k = 1; : : : ; N do
if Light to Colors then
ILab ...
34/51
Introduction Our method Summary
Algorithm
Chromatic adaptation
for k = 1; : : : ; N do
if Light to Colors then
ILab ...
lter
 Reference illuminant estimation:
Gray World assumption
[FSDP14, HCWxW06]
 Local CAT algorithm [KJF07]
 Avoids undesi...
35/51
Introduction Our method Summary
Evaluation
Outline
1 Introduction
Context
Objective
Contributions
2 Our method
Algor...
36/51
Introduction Our method Summary
Evaluation
Subjective evaluation
 15 users
 Results from 4
state-of-the-arts
 Result...
37/51
Introduction Our method Summary
Evaluation
Subjective evaluation: the analysis
Ours Pitie Reinhard Bonneel Tai
2345
...
38/51
Introduction Our method Summary
Evaluation
Objective evaluation
SSIM
It measures the degree of artifacts in the resu...
Prochain SlideShare
Chargement dans…5
×

Style-aware robust color transfer

762 vues

Publié le

Style-aware robust color transfer, H. Hristova, O. Le Meur, R. Cozot and K. Bouatouch, Computational Aesthetic, 2015.

Transferring features, such as light and colors, between input and reference images is the main objective of color transfer methods. Current state-of-the-art methods focus mainly on the complete transfer of the light and color distributions. However, they do not successfully grasp specific light and color variations in image styles. In this paper, we propose a local method for carrying out a transfer of style between two images. Our method partitions both images to Gaussian distributed clusters by considering their main style features. These features are automatically determined by the classification step of our algorithm. Moreover, we present several novel policies for input/reference cluster mapping, which have not been tackled so far by previous methods. To complete the style transfer, for each pair of corresponding clusters, we apply a parametric color transfer method and a local chromatic adaptation transform. Results, subjective user evaluation as well as objective evaluation show that the proposed method obtains visually pleasing and artifact-free images, respecting the reference style.

Publié dans : Technologie
  • Soyez le premier à commenter

  • Soyez le premier à aimer ceci

Style-aware robust color transfer

  1. 1. 1/51 Introduction Our method Summary Style-aware robust color transfer H. Hristova O. Le Meur R. Cozot K. Bouatouch IRISA - University of Rennes 1 Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  2. 2. 2/51 Introduction Our method Summary Outline 1 Introduction Context Objective Contributions 2 Our method Algorithm Evaluation Results 3 Summary Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  3. 3. 3/51 Introduction Our method Summary Context Outline 1 Introduction Context Objective Contributions 2 Our method Algorithm Evaluation Results 3 Summary Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  4. 4. 4/51 Introduction Our method Summary Context Color transfer domain Objective of color transfer Color transfer aims at modifying the look of an input image considering the illumination and the color palette of a reference image (e.g., considering statistical characteristics of the light and color distributions, histograms, gradients, etc.). Input Our result Reference Transformation f(µ, Σ, φ, ...) light, colors light, colors Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  5. 5. 5/51 Introduction Our method Summary Context State-of-the-art global methods [Reinhard et al., 2001] Linear mapping L
  6. 6. Diagonal covariance matrix Gaussian distribution [Pitie et al., 2007] Linear mapping as a solution to Monge-Kantorovich optimization problem CIE Lab Non-diagonal covariance matrix Gaussian distribution Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  7. 7. 6/51 Introduction Our method Summary Context State-of-the-art local methods [Tai et al., 2005] Linear mapping: [Reinhard et al., 2001] L
  8. 8. 3D image clustering Cluster mapping: maps the closest clusters in terms of their luminance channel values Gaussian distribution (cluster-wise) [Bonneel et al., 2013] Linear mapping: [Pitie et al., 2007] CIE Lab Image clustering in terms of the luminance channel Cluster mapping: shadows to shadows, midtones to midtones, highlights to highlights Gaussian distribution (cluster-wise) Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  9. 9. 7/51 Introduction Our method Summary Context Limitations Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  10. 10. 7/51 Introduction Our method Summary Context Limitations Style inconsistency Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  11. 11. 7/51 Introduction Our method Summary Context Limitations Style inconsistency InputimageReferenceimage[Reinhardetal.2001] Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  12. 12. 8/51 Introduction Our method Summary Context Limitations Style inconsistency Gaussian distribution Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  13. 13. 8/51 Introduction Our method Summary Context Limitations Style inconsistency Gaussian distribution Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  14. 14. 9/51 Introduction Our method Summary Context Limitations Style inconsistency Gaussian distribution Clustering on the same color space components for input/reference images Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  15. 15. 9/51 Introduction Our method Summary Context Limitations Style inconsistency Gaussian distribution Clustering on the same color space components for input/reference images Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  16. 16. 10/51 Introduction Our method Summary Context Limitations Style inconsistency Gaussian distribution Clustering on the same color space components for input/reference images Over-saturation or lack of contrast Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  17. 17. 10/51 Introduction Our method Summary Context Limitations Style inconsistency Gaussian distribution Clustering on the same color space components for input/reference images Over-saturation or lack of contrast [Pitie et al., 2007] [Reinhard et al., 2001] Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  18. 18. 11/51 Introduction Our method Summary Context Limitations Style inconsistency Gaussian distribution Clustering on the same color space components for input/reference images Over-saturation or lack of contrast Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  19. 19. 12/51 Introduction Our method Summary Objective Outline 1 Introduction Context Objective Contributions 2 Our method Algorithm Evaluation Results 3 Summary Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  20. 20. 13/51 Introduction Our method Summary Objective Our objective Transferring the light and color features of a reference image to an input image with respect to the reference style features. Obtaining a naturally looking image which style features are as close as possible to those of the reference style. Style notion in our context In our method, style in images is characterized by two features: color and light. Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  21. 21. 14/51 Introduction Our method Summary Objective Example Input Reference Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  22. 22. 15/51 Introduction Our method Summary Objective Example Input Reference [Pitieetal.,2007] Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  23. 23. 16/51 Introduction Our method Summary Objective Example Input Reference [Pitieetal.,2007][Bonneeletal.,2013] Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  24. 24. 17/51 Introduction Our method Summary Objective Example Input Reference [Pitieetal.,2007][Bonneeletal.,2013] Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  25. 25. 18/51 Introduction Our method Summary Contributions Outline 1 Introduction Context Objective Contributions 2 Our method Algorithm Evaluation Results 3 Summary Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  26. 26. 19/51 Introduction Our method Summary Contributions Classi
  27. 27. cation of the input and reference images Light-based style images Colors-based style images Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  28. 28. 19/51 Introduction Our method Summary Contributions Classi
  29. 29. cation of the input and reference images Light-based style images Colors-based style images Two ways of clustering: Luminance channel Luminance-Hue distributions Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  30. 30. 19/51 Introduction Our method Summary Contributions Classi
  31. 31. cation of the input and reference images Light-based style images Colors-based style images Two ways of clustering: Luminance channel Luminance-Hue distributions Four mapping policies: Light to Light, Light to Colors, Colors to Light, Colors to Colors Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  32. 32. 19/51 Introduction Our method Summary Contributions Classi
  33. 33. cation of the input and reference images Light-based style images Colors-based style images Two ways of clustering: Luminance channel Luminance-Hue distributions Four mapping policies: Light to Light, Light to Colors, Colors to Light, Colors to Colors Evaluation of the results Subjective user study Objective metrics Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  34. 34. 20/51 Introduction Our method Summary Algorithm Outline 1 Introduction Context Objective Contributions 2 Our method Algorithm Evaluation Results 3 Summary Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  35. 35. 21/51 Introduction Our method Summary Algorithm Framework light, colors light, colors Input image Reference image Classification Clustering Clustering Chromatic adaptation Color transfer Mergingprocess Mapping policies Input clusters Set of corresponding clusters Final image Reference clusters Classification Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  36. 36. 22/51 Introduction Our method Summary Algorithm Automatic image classi
  37. 37. cation system Colors-based style images Images which color information is sucient enough to well-de
  38. 38. ne at least two dierent and signi
  39. 39. cant colors. Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  40. 40. 23/51 Introduction Our method Summary Algorithm Automatic image classi
  41. 41. cation system Light-based style images Images which are not classi
  42. 42. ed as colors-based style images are classi
  43. 43. ed as light-based style images. Their light features are more meaningful than their color features. Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  44. 44. 24/51 Introduction Our method Summary Algorithm Framework light, colors light, colors Input image Reference image Classification Clustering Clustering Chromatic adaptation Color transfer Mergingprocess Mapping policies Input clusters Set of corresponding clusters Final image Reference clusters Classification Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  45. 45. 25/51 Introduction Our method Summary Algorithm Clustering: Gaussian mixture models Light-based style image Colors-based style image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  46. 46. 26/51 Introduction Our method Summary Algorithm Framework light, colors light, colors Input image Reference image Classification Clustering Clustering Chromatic adaptation Color transfer Mergingprocess Mapping policies Input clusters Set of corresponding clusters Final image Reference clusters Classification Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  47. 47. 27/51 Introduction Our method Summary Algorithm Pseudo code algorithm for k = 1; : : : ; N do if Light to Colors then ILab := FindDarkestCluster() JLab := FindColdestCluster() end if if Colors to Light then ILab := FindColdestCluster() JLab := FindDarkestCluster() end if if Light to Light then ILab := FindDarkestCluster() JLab := FindDarkestCluster() end if if Colors to Colors then [ILab, JLab] := FindMinDistPair() end if OLab := PerformTranspOnAB(ILab, JLab) Exclude ILch Exclude JLch end for Ofinal rgb := CATLocal(Orgb; Jrgb) Darkest cluster - the cluster which centroid has the minimum luminance value Coldest cluster - the cluster which centroid has the maximum hue value Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  48. 48. 28/51 Introduction Our method Summary Algorithm Mapping policies for k = 1; : : : ; N do if Light to Colors then ILab := FindDarkestCluster() JLab := FindColdestCluster() end if if Colors to Light then ILab := FindColdestCluster() JLab := FindDarkestCluster() end if if Light to Light then ILab := FindDarkestCluster() JLab := FindDarkestCluster() end if if Colors to Colors then [ILab, JLab] := FindMinDistPair() end if OLab := PerformTranspOnAB(ILab, JLab) Exclude ILch Exclude JLch end for Ofinal rgb := CATLocal(Orgb; Jrgb) Darkest cluster - the cluster which centroid has the minimum luminance value Coldest cluster - the cluster which centroid has the maximum hue value Light to Colors Input light-based style image Reference colors-based style image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  49. 49. 29/51 Introduction Our method Summary Algorithm Mapping policies for k = 1; : : : ; N do if Light to Colors then ILab := FindDarkestCluster() JLab := FindColdestCluster() end if if Colors to Light then ILab := FindColdestCluster() JLab := FindDarkestCluster() end if if Light to Light then ILab := FindDarkestCluster() JLab := FindDarkestCluster() end if if Colors to Colors then [ILab, JLab] := FindMinDistPair() end if OLab := PerformTranspOnAB(ILab, JLab) Exclude ILch Exclude JLch end for Ofinal rgb := CATLocal(Orgb; Jrgb) Darkest cluster - the cluster which centroid has the minimum luminance value Coldest cluster - the cluster which centroid has the maximum hue value Colors to Light Input colors-based style image Reference light-based style image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  50. 50. 30/51 Introduction Our method Summary Algorithm Mapping policies for k = 1; : : : ; N do if Light to Colors then ILab := FindDarkestCluster() JLab := FindColdestCluster() end if if Colors to Light then ILab := FindColdestCluster() JLab := FindDarkestCluster() end if if Light to Light then ILab := FindDarkestCluster() JLab := FindDarkestCluster() end if if Colors to Colors then [ILab, JLab] := FindMinDistPair() end if OLab := PerformTranspOnAB(ILab, JLab) Exclude ILch Exclude JLch end for Ofinal rgb := CATLocal(Orgb; Jrgb) Darkest cluster - the cluster which centroid has the minimum luminance value Coldest cluster - the cluster which centroid has the maximum hue value Light to Light Input light-based style image Reference light-based style image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  51. 51. 31/51 Introduction Our method Summary Algorithm Mapping policies for k = 1; : : : ; N do if Light to Colors then ILab := FindDarkestCluster() JLab := FindColdestCluster() end if if Colors to Light then ILab := FindColdestCluster() JLab := FindDarkestCluster() end if if Light to Light then ILab := FindDarkestCluster() JLab := FindDarkestCluster() end if if Colors to Colors then [ILab, JLab] := FindMinDistPair() end if OLab = PerformTranspOnAB(ILab, JLab) Exclude ILch Exclude JLch end for Ofinal rgb := CATLocal(Orgb; Jrgb) Darkest cluster - the cluster which centroid has the minimum luminance value Coldest cluster - the cluster which centroid has the maximum hue value Colors to Colors Input colors-based style image Reference colors-based style image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  52. 52. 32/51 Introduction Our method Summary Algorithm Color transformation for k = 1; : : : ; N do if Light to Colors then ILab := FindDarkestCluster() JLab := FindColdestCluster() end if if Colors to Light then ILab := FindColdestCluster() JLab := FindDarkestCluster() end if if Light to Light then ILab := FindDarkestCluster() JLab := FindDarkestCluster() end if if Colors to Colors then [ILab, JLab] := FindMinDistPair() end if OLab := PerformTranspOnAB(ILab, JLab) Exclude ILch Exclude JLch end for Ofinal rgb := CATLocal(Orgb; Jrgb) Color transformation between each two corresponding clusters Carried out on the a and b channels of CIE Lab Closed-form solution: solution to Monge-Kantorovich's optimization problem Non-diagonal covariance matrix between the channels of the color space Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  53. 53. 33/51 Introduction Our method Summary Algorithm Color transformation for k = 1; : : : ; N do if Light to Colors then ILab := FindDarkestCluster() JLab := FindColdestCluster() end if if Colors to Light then ILab := FindColdestCluster() JLab := FindDarkestCluster() end if if Light to Light then ILab := FindDarkestCluster() JLab := FindDarkestCluster() end if if Colors to Colors then [ILab, JLab] := FindMinDistPair() end if OLab := PerformTranspOnAB(ILab, JLab) Exclude ILch Exclude JLch end for Ofinal rgb := CATLocal(Orgb; Jrgb) Color transformation between each two corresponding clusters Carried out on the a and b channels of CIE Lab Closed-form solution: solution to Monge-Kantorovich optimization problem Non-diagonal covariance matrix between the channels of the color space Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  54. 54. 34/51 Introduction Our method Summary Algorithm Chromatic adaptation for k = 1; : : : ; N do if Light to Colors then ILab := FindDarkestCluster() JLab := FindColdestCluster() end if if Colors to Light then ILab := FindColdestCluster() JLab := FindDarkestCluster() end if if Light to Light then ILab := FindDarkestCluster() JLab := FindDarkestCluster() end if if Colors to Colors then [ILab, JLab] := FindMinDistPair() end if OLab := PerformTranspOnAB(ILab, JLab) Exclude ILch Exclude JLch end for Ofinal rgb := CATLocal(Orgb; Jrgb) Adapting pixel-wisely the colors of the output to the reference illuminant Input illuminant: white image obtained by performing Gaussian low-pass
  55. 55. lter Reference illuminant estimation: Gray World assumption [FSDP14, HCWxW06] Local CAT algorithm [KJF07] Avoids undesired color saturation, prevents the result from becoming at and reduces false colors in the result Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  56. 56. 35/51 Introduction Our method Summary Evaluation Outline 1 Introduction Context Objective Contributions 2 Our method Algorithm Evaluation Results 3 Summary Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  57. 57. 36/51 Introduction Our method Summary Evaluation Subjective evaluation 15 users Results from 4 state-of-the-arts Results from our method Style match evaluation Aesthetic pleasingness evaluation 5-point scale Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  58. 58. 37/51 Introduction Our method Summary Evaluation Subjective evaluation: the analysis Ours Pitie Reinhard Bonneel Tai 2345 Ours Pitie Reinhard Bonneel Tai Stylescore * * *** *** Ours Pitie Reinhard Bonneel Tai 12345 Ours Pitie Reinhard Bonneel TaiAestheticsscore ** *** * Box-and-Whisker plots Paired T-tests Correlation between the two scores Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  59. 59. 38/51 Introduction Our method Summary Evaluation Objective evaluation SSIM It measures the degree of artifacts in the result. It is applied between the input image and the result. S(x; y) = f (c(x; y); s(x; y)) (1) The luminance component is removed from the computation of the metrics leaving the structural and contrast components. Bhattacharya coecient Measures the distance between the result/reference histograms [ATR98] of the components of CIE Lab color space. Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  60. 60. 39/51 Introduction Our method Summary Evaluation Objective evaluation: the analysis (a) Our method (b) Pitié et al. (c) Bonneel at al.(d) Reinhard et al. (e) Tai et al. SSIM Bhattacharya 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 8 9 x 10 −3 SSIM Bhattacharya 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 x 10 −3 SSIM Bhattacharya 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 x 10 −3 SSIM Bhattacharya 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 x 10 −3 SSIM Bhattacharya 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 x 10 −3 Joint distribution of SSIM and Bhattacharya coecient Paired T-tests (SSIM, Bhattacharya) = (1, 1): optimal for the pair Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  61. 61. 40/51 Introduction Our method Summary Results Outline 1 Introduction Context Objective Contributions 2 Our method Algorithm Evaluation Results 3 Summary Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  62. 62. 41/51 Introduction Our method Summary Results Results: Light to Colors Result Input image Reference image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  63. 63. 42/51 Introduction Our method Summary Results Results: Colors to Light Input image Reference image Result Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  64. 64. 43/51 Introduction Our method Summary Results Results: Light to Light Result Input image Reference image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  65. 65. 44/51 Introduction Our method Summary Results Results: Colors to Colors Result Input image Reference image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  66. 66. 45/51 Introduction Our method Summary Results Results: classical example Result Input image Reference image Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  67. 67. 46/51 Introduction Our method Summary Summary New method for style transfer for a wide class of image pairs Automatic image classi
  68. 68. cation Four mappings policies Subjective and objective evaluations Future work Enriching the image feature space Finding a connection between the subjective user study and the objective evaluation Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  69. 69. 47/51 Introduction Our method Summary Thank you for your attention! More results and information on http: // people. irisa. fr/ Hristina. Hristova Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  70. 70. 48/51 Appendix References References I Aherne F. J., Thacker N. A., Rockett P. I.: The bhattacharyya metric as an absolute similarity measure for frequency coded data. Kybernetika 34, 4 (1998), 363{368. Bonneel N., Sunkavalli K., Paris S., Pfister H.: Example-based video color grading. ACM Transactions on Graphics (Proceedings of SIGGRAPH 2013) 32, 4 (2013), 2. Frigo O., Sabater N., Demoulin V., Pierre H.: Optimal transportation for example-guided color transfer. 12th Asian Conference on Computer Vision (ACCV) (2014). Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  71. 71. 49/51 Appendix References References II Huo J., Chang Y., Wang J., xia Wei X.: Robust automatic white balance algorithm using gray color points in images. IEEE Trans. Consumer Electronics 52, 2 (2006), 541{546. URL: http://dblp.uni-trier.de/db/journals/tce/ tce52.html#HuoCWW06. Kuang J., Johnson G. M., Fairchild M. D.: icam06: A re
  72. 72. ned image appearance model for hdr image rendering. Journal of Visual Communication and Image Representation 18, 5 (2007), 406{414. Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  73. 73. 50/51 Appendix References References III Pitie F., Kokaram A. C., Dahyot R.: Automated colour grading using colour distribution transfer. Computer Vision and Image Understanding 107, 1 (2007), 123{137. Reinhard E., Ashikhmin M., Gooch B., Shirley P.: Color transfer between images. Computer Graphics and Applications, IEEE 21, 5 (2001), 34{41. Shih Y., Paris S., Durand F., Freeman W. T.: Data-driven hallucination of dierent times of day from a single outdoor photo. ACM Transactions on Graphics (TOG) 32, 6 (2013), 200. Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch
  74. 74. 51/51 Appendix References References IV Tai Y.-W., Jia J., Tang C.-K.: Local color transfer via probabilistic segmentation by expectation-maximization. In Computer Vision and Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society Conference on (2005), vol. 1, IEEE, pp. 747{754. Style-aware robust color transfer H. Hristova, O. Le Meur, R. Cozot, K. Bouatouch

×