Predicting visible image degradation by colour image difference formulae

0
335

The article titled “Predicting visible image degradation by colour image difference formulae” written by Eriko Bando, Jon Y. Hardeberg, David Connah, and Ivar Farup in 2004, investigates the relationship between color difference metrics in digital images and human perception of those differences. In the context of the increasing use of digital images and the diversity of image reproduction methods, this study focuses on evaluating the ability of different color difference formulas to visually predict the degradation of reproduced images.

Traditionally, the CIELAB color difference equation ( \Delta E_{ab} ) has been widely used in the graphic industry to measure the color difference between pairs of color patches. However, when it comes to digital images, this formula has proven insufficient to predict perceived difference due to its inability to consider the spatial sensitivity of the human eye. To mitigate this limitation, image difference metrics incorporating more complex models of the human visual system have been developed.

The study uses three color difference metrics: the traditional CIELAB formula ( \Delta E_{ab} ), the S-CIELAB metric, and the extended image appearance model iCAM. While S-CIELAB is an extension of CIELAB designed to account for the spatial sensitivity of the human eye, the iCAM model is based on similar spatial vision concepts but includes more sophisticated models of chromatic adaptation.

To evaluate these metrics, a psychophysical experiment was conducted based on a CRT monitor, using six original images reproduced by six different color gamut mapping algorithms. The reproduced images were evaluated by a total of 16 observers, who compared pairs of images (one original and one reproduced) and selected which showed the least perceptual difference. The experimentation was carried out under controlled viewing conditions to ensure consistent color reproduction.

The results obtained reveal that the color differences calculated by each metric do not directly correlate with the differences perceived by the observers. Specifically, it was found that neither S-CIELAB nor iCAM accurately predicted the perceived differences between images. For example, although the gamut mapping algorithm that preserves hue angle through minimal clipping of ( \Delta E_{ab} ) showed good overall performance, subjective evaluations varied considerably depending on the algorithm and the specific image.

A critical observation was that perceived color difference largely depends on the image background. Images with uniform color backgrounds tended to show atypical results compared to images with complex textures. This finding suggests that the structure of the image background may have a significant impact on human perception of image degradation.

In conclusion, the study emphasizes the need to continue researching to improve color difference formulas, such as iCAM and S-CIELAB, and to develop new methods that take into account factors like the image background and other contextual elements. The results suggest that perception of image quality cannot be predicted solely from computational differences calculated at the pixel level, highlighting the complexity of capturing the human visual experience through objective metrics.

Review on the publisher’s website