Color Transfer between Images

Download Report

Transcript Color Transfer between Images

Color Transfer between
Images
Reporter:Chen Shuangmin
CAGD&CG
Introduction to the authors
• Erik Reinhard
.is a postdoctoral fellow at the University of
Utah
.interested in visual-perception and
parllelgraphics fields.
.BS diploma in computer science from Deft
University of Technology and a phD in
computer science from the university of
Bristol
CAGD&CG
Introduction to the authors
• Bruce Gooch
.is a graduate student at the university of
Utah
.intersted in nonphotorealistic rendering
.BS in mathematics and an MS in
computer science from the university of
Utah
CAGD&CG
Introduction to the authors
• Michael Ashikhmin
.an assistant professor at State University
of New York
.intersted in developing simple and
practical algorithms for computer graphics
.has an MS in physics from the Moscow
Institute of Physics and Technology,an MS
in chemistry from the University of
California
CAGD&CG
Guidence
• Color correction(remove undesired colorillumination) - in one image
Red eye
restoration
CAGD&CG
Guidence
• Color correction(remove undesired colorillumination)
• Borrow color from one image to another
source
target
Similar composition
CAGD&CG
Main idea
1.Color constancy correction
.global statistics
.swatches
2.Gamma correction(logxr = rlogx)
Goal
• Alter an image’s color
CAGD&CG
1.Color constancy correction
--- Gloabal statistics
• Why choose lab color space???
.RGB
.lab
.CIECAM97s
• Why decorrelation is better???
CAGD&CG
Introduction to RGB
• We can get any color by blending with
RGB
Definition with cubic
CAGD&CG
Introduction to lab color space
• In Photoshop,the range of l is 0 to 100;and
in color palete,the ranges of a and b are
both 120 to -120.
It will be more easily
understood in Gamma
correction
CAGD&CG
Why choose lab color space?
--decorrelation
2000 random points
CAGD&CG
Why decorrelation is better?
• Apply different operations in different color
channels that without undesirable
crosschannel artifacts.
• The data in lab space is more compressed
(log space)
• RGB:correlations between the different
channel’s values. (0,0,0) - (255,255,255)
CAGD&CG
Comparison of the effects
In three color spaces
Source image
Target image
CAGD&CG
Comparison of effects
In three color spaces
RGB color space
Lab color space
CIECAM97s
CAGD&CG
Convertion between color spaces
• RGB
.RGB
.XYZ
.LMS
.LMS
• lab
lab
XYZ (international standard)
LMS
LMS
lab
RGB
CAGD&CG
About CIECAM97s
RGB XYZ (international standard)
XYZ
LMS
CAGD&CG
Convertion between color spaces
• RGB
.RGB
.XYZ
lab
XYZ
LMS
CAGD&CG
Convertion between color spaces
• RGB
.LMS
.LMS
lab
LMS
lab
CAGD&CG
Convertion between color spaces
• RGB
• lab
lab
RGB(inverse operations)
L = 10L
M = 10M
S = 10S
CAGD&CG
Main idea
1.Color constancy correction
.global statistics
.swatches
2.Gamma correction(logxr = rlogx)
CAGD&CG
Two concept in statistics
• Standard deviation
• Mean value
Show with chalk on
blackboard
CAGD&CG
Color constancy correction
• Subtract the mean from the data points
• Scale the data points comprising the
synthetic image by factors determined by
the respective standard deviation
CAGD&CG
Color constancy correction
• add the averages computed for the
photagraph.
+ <lt>
+ <at>
+ <bt>
• Displaying: convert the result back to RGB
via logLMS,LMS,and XYZ color spaces
using equations
CAGD&CG
Convert back to RGB for displaying
• RGB
• lab
lab
RGB(inverse operations)
L = 10L
M = 10M
S = 10S
CAGD&CG
Main idea
1.Color constancy correction
.global statistics
.swatches
2.Gamma correction(logxr = rlogx)
CAGD&CG
Swatches
• When source and target images don’t
work well together – swatches.
CAGD&CG
Swatches
• Several pairs of clusters in lab space
(cluster for each swatch)
- Compute statistics for each swatch
CAGD&CG
Swatches -result
• When source and target images don’t
work well – swatches.
target
source
result
CAGD&CG
Main idea
1.Color constancy correction
.global statistics
.swatches
2.Gamma correction(logxr = r logx)
CAGD&CG
My question about this paper
• I don’t know how to decide which is source
image and which is target image. It will
give different effects.
T
result
T
S
result
S
I don’t like this result!
This effect is much better!
I like it.
CAGD&CG
My question about this paper
• I don’t know which two images will be
matching better .How to choose?
result
result
This effect is not much better!
This effect is much better!
I like it.(result of this paper.)
CAGD&CG
My question about this paper
• The size of these two images is related to
the result.
Target 800*600
Source 426*266
Target 454*366
Source 426*266
CAGD&CG
Thank you!
CAGD&CG