Perceptual Evaluation of Colour Gamut Mapping Algorithms (.vnd.ms-powerpoint)

Download Report

Transcript Perceptual Evaluation of Colour Gamut Mapping Algorithms (.vnd.ms-powerpoint)

Perceptual Evaluation of Colour
Gamut Mapping Algorithms
Fabienne Dugay
The Norwegian Color Research Laboratory
Faculty of Computer Science and Media Technology
Gjøvik University College, Gjøvik, Norway
[email protected]
http://www.colorlab.no
Master’s thesis presentation, 7th June 2007
2
Outline

Introduction



Gamut mapping algorithms




Gamut mapping algorithms (GMAs)
Experimental setup


Colour Gamuts
Goal
Psychophysical evaluation
Images, Media, Viewing conditions
Results & Analysis
Conclusion and perspectives
3
Introduction

Gamut = range of reproducible colours of a device or range of
colours in a image

Printers have smaller gamut than monitor


How to reproduce those out-of-gamut colours ?
Gamut mapping algorithms (GMAs): ensure a good
correspondence of overall colour appearance between the
original and the reproduction
4
Goal

Evaluate the performance of selected GMAs on real images

Influence of the test images

Influence of the observers

Influence of the experiments
5
Gamut mapping algorithms

Non-spatial GMAs



The image is treated globally
Gamut compression or gamut clipping
Spatial GMAs


Depend on the neighbourhood pixels
Balance both colour accuracy and preservation of details
6
Experimental methods

No metrics have been proved to be efficient for evaluating the
performance of GMAs

Psychophysical tests with a panel of observers

20 observers (11 “experts” & 9 “non-experts”)

Asked about the accuracy of the reproductions

The raw data from the experiments are treated statistically to
obtain z-scores
7
Experimental methods


20 test images with various characteristics
Original: sRGB image on calibrated monitor
8
Experimental methods

Reproductions on a inkjet printer with plain paper
9
Experimental methods

5 GMAs:

HPminDE: Hue preserving minimum delta E clipping
SGCK: lightness and chroma compression, hue preserving




Zolliker: recovers local contrast, preserves lightness and
saturation
Kolås: hue and edge preserving spatial GMA
Gatta: preserves hue and local relationships
10
Experimental methods

Viewing conditions follow the CIE guidelines:





Simulated D50 lights for the prints
D65 white point for the monitor
Viewed in a neutral grey room with lights at their minimum
intensity
Original and reproduction images have the same size and a white
border
Neutral grey background
11
Experimental methods

Two psychophysical experiments

With printed reproductions


Ranking (rank the 5 reproductions from the most to the least accurate
to the original displayed on the monitor)
With simulated printed reproductions on screen

Pair comparison (choose the most accurate reproduction in a pair)
12
Results

Results from the ranking experiment
13
Analysis

HPminDE: not an accurate GMA

Kolås, SGCK and Gatta not significantly different

A spatial and non-spatial GMAs seen as accurate
14
Results

Results from the ranking experiment, for each image and GMA
15
Analysis

Dependant on the test images
16
Analysis

But strong correlation between the % of out-of-gamut colours
and the number of distinguishable GMAs

Strong correlation between the % of out-of-gamut colours and
the perceived difficulty to rank the reproductions

Gamut mapping especially important when dealing with small
gamut devices
17
Results
Dependant on the observers
Accuracy scores for all images, by experts and non experts
0.8
0.6
0.4
0.2
0
HPminDE
Accuracy

SGCK
Zolliker
-0.2
Kolaas
Gatta
Experts
Non-experts
-0.4
-0.6
-0.8
-1
-1.2
-1.4
GMAs
18
Analysis





Different results between the two groups
Stronger consensus among the experts
All GMAs have tight scores for the non-experts
Experts look at the best rendering of details
Non-experts look more at the saturation
19
Results
Dependant on the experiments
Comparison paper and screen, all observers
0.4
0.2
0
Accuracy

HPminDE
SGCK
Zolliker
Kolaas
Gatta
Paper
Screen
-0.2
-0.4
-0.6
-0.8
GMAs
20
Analysis

Globally comparable results

Some other parameters:




Random of the scenes
Accuracy or preference?
Other media/printers
LCD/CRT monitors
21
Conclusion and perspectives

None GMA is significantly better than all the others

HPminDE (clipping) is not perceived as an accurate GMA

The choice of a efficient GMA may depend on the image, the
media, the target customer and an universal GMA seems
inexistent

Meta-analysis to join the results of the different GMA
evaluations?
22
Thank you for you attention
Any questions?