Evaluating CATs as Predictors of Observer Adjustments in Softcopy Fine Art Reproduction |
 | |
A project to evaluate current practices in fine art image reproduction is conducted in which pieces of artwork in various media are being imaged by participating museums. As part of this project, observers were asked to make adjustments in softcopy fine art reproductions. The goal is to see how people working in museums, libraries and archives make color adjustments to artwork presented to them on screen. Observers were led through various interfaces allowing them to adjust the image seen on the screen to better represent the original in a light booth. They were asked to make adjustments until the screen image was good enough, as an exact match may neither be possible nor necessary for us to detect relevant patterns in adjustment among observers. Patterns or trends in adjustments by observers can be used as an indication of how images should be processed to match with the adjustment by observers most closely. The adjustments by observers were compared with the prediction by three chromatic adaptation models. Overall the Fairchild92 model outperforms the Bradford and CAT02 transformations in matching with adjustments by observers more closely.
|
Publications
Jun Jiang, Franziska Frey, and Susan Farnand.
Evaluating CATs as Predictors of Observer Adjustments in Softcopy Fine Art Reproduction. Color and Imaging Conference (CIC) 2010.
|
Images
 |
|
Experimental setup:
A 27 inch Apple Cinema Display is used for showing softcopy reproductions. Observers were asked to adjust the images on the display to match with the original hard copy in the light booth.
|
 |
|
User interface for image adjustment:
Hue adjustments (a) allows users to change the hue of the image in different hue directions. Global adjustments (b) allows changes in image brightness, contrast, saturation, and sharpness. The global adjustments are indiscriminate to colors, while local adjustments (c) can be used in order to make certain colors right without affecting other colors in the image. The user interfaces are designed and implemented in Matlab and Psychtoolbox. C code is used reflect the changes in images in real time.
|
 |
|
Predictions of adjusted images by observers using CAT models:
The original image is shown in (a). The adjusted image by one observer is shown in (b), and it matches with the predictions by three chromatic adaptation models (c), (d) and (e) much more closely than with the starting image (a).
|
 |
|
Error map between the adjusted images by observers and the predictions by CAT models:
The error distribution map of 'Daisy' (after the S-CIELAB model) is shown between the adjusted image by one observer and the predictions by three models.
Three chromatic adaptation transforms (CATs), Bradford, Fairchild92 and CAT02 are selected to predict adjustments by observers. Bradford transformation is essentially a von Kries transformation with an additional exponential nonlinearity on the blue channel. In the experiment, the linearized Bradford transformation is used. The CAT02 and Fairchild92 models are linear in nature and both can predict incomplete adaptation.
|
 |
|
Analysis of variance (ANOVA) :
ANOVA is used to analyze the factors (image, CAT model, ...) that may contribute to the image difference between the adjusted images by observers and the predictions by three CAT models. The y-axis is the mean image difference. The smaller the image difference, the better the model is.
No chromatic adaptation model is found to better predict adjustments by observers across all images used in the experiment. The Fairchild92 model generally matches with the visual editing by observers more closely than the Bradford or CAT02 model except for images with near-neutral appearance.
|
|
Slides
Color and Imaging Conference (CIC) 2010 poster
|
Current Practices in Fine Art Reproduction
|
|