Evaluating the Perceived Quality of Soft-copy Reproductions of Fine Art Images With and Without the Original Present |
 | |
A project to evaluate the perceived image quality of fine art reproductions was conducted in which pieces of artwork were imaged by participating institutions. As part of this project, observers were asked to rank the fine art reproductions on a characterized display either with or without the presence of the originals. The goal was to see whether the availability of the original artwork influenced how people appreciated the reproductions. A low correlation was found between the ranking results with and without the originals, indicating a shift in the criterion employed by observers in evaluating perceived image quality from color accuracy to preference. A web-based experiment was designed in order to better understand the necessity of a controlled environment to ensure image quality evaluated based on preference. A significantly high correlation was found between the results from the experiment without the original and from the one conducted online. Therefore, the preference judgments of perceived image quality were stable regardless of changes in viewing conditions. In addition, the areas that were considered most important by observers in making ranking decisions were identifiable by observers’ clicks on the image in the web-based experiment. By understanding the part of the paintings to which more attention was drawn, information regarding the image saliency could be learned.
|
Publications
Jun Jiang, Franziska Frey and Susan Farnand.
Evaluating the Perceived Quality of Soft-copy Reproductions of Fine Art Images With and Without the Original Present. Color and Imaging Conference (CIC), 2011.
|
Images
 |
|
Display characterization to ensure accurate color reproduction:
An LMT 1210 colorimeter is used to characterize the display to ensure accurate color reproduction. LMT Colorimeter is chosen because of its high sensitivity at low light level. The display characterization model by Day, Taplin and Berns is used to map between LCD digital counts and XYZ tristimulus values.
The colorimetric performance of the display (Apple 27 inch Cinema Display) is reasonably good after characterization. The mean and max color differences (CIEDE2010) are 1.4 and 2.2, respectively.
|
 |
|
Experimental setup to evaluate the perceived image quality:
Both the soft copy images on the display and the original artwork in the light booth can be viewed side-by-side simultaneously by observers. During the experiment, observers were instructed to click on the image on the display that represented the original better. Software interfaces were developed in Matlab on the extension provided by the high-level Psychophysics Toolbox.
|
 |
|
Web-based experiment to evaluate the perceived image quality:
A web application is used to evaluate fine art reproductions in an uncontrolled environment by observers from almost anywhere with little constraints on test conditions, as long as reasonable Internet speed and a web browser are available.
|
 |
|
User clicks on 'Firelight' that most contribute to image quality evaluation:
In the web-based experiment, observers are asked not only to pick their preferred image, but also to click in the area of the image that most influence their decisions.
User clicks are highly clustered on the cheek of the lady's face, indicating the importance of faithful reproduction of the color of human skin.
|
 |
|
The correlation between three experiments:
Ranking experiments are conducted under three different conditions. The first two are performed on a characterized display in Munsell Color Science Laboratory (MCSL). The main difference between the first two is that the original artwork is available in the first experiment but not in the second one. The third experiment is run online, so that data can be gathered in an uncontrolled environment.
A high correlation is observed between the ranks of images in the web-based experiment and that w/o the original in the lab. On the other hand, in the plot on the left, little correlation is found between the ranking results in the web experiment and that w/ the original in the lab environment. Similar results are found for the rest of the paintings. Given the high correlation between the two experiments w/o the original, a similar criterion, image preference, is likely to be shared by observers when the images are evaluated (whether the viewing condition is controlled or not).
|
 |
|
Latent structure of test images:
Clustering analysis is used to group the paintings based on how they are ranked in image quality. The grouping of images for the experiment w/o the original in MCSL is the same as the web-based experiment (the plot on the right).
'Daisy' is ranked closer to neutral color images when the original is available, but more similarly to colorful images when the original is absent.
|
|
Slides
Color and Imaging Conference 2011 Presentation
|
Current Practices in Fine Art Reproduction
|
|