Typically when we begin a career in computational color science, we start from colorimetric color reproduction. As we explore the wide variety of device gamuts, we quickly progress to perceptual color reproduction. At the latest when we build shippable product, we have to progress to preferred color reproduction.
For most, this is the goal. Only few braves—like for example Larry Lavendel and Tim Kohler in Canon's Color Advisor—venture in the bold land of pleasing color, because it actually involves heavily editing the hues in an image.
The current issue of the Journal of Electronic Imaging has a paper on a new algorithm for achieving pleasing color: Zhen Tang, Zhenjiang Miao, Yanli Wan and Zhifei Wang, "Color harmonization for images", J. Electron. Imaging 20, 023001 (Apr 13, 2011); doi:10.1117/1.3574097.
Of course, the algorithm has a high computational complexity and the authors make use of the GPU. In view of this, it puzzles me why they are using the H coordinate of the HSV (hue, saturation, value) color space as a correlate for hue, because its coordinates are not even well decorrelated, not to mention that the correlation with perceptual quantities is rather poor. For example, yellow and blue have the same V value in HSV.
Today, one should have a good reason not use use a hue correlate from a color appearance model like CIECAM. I wonder if such a choice would have solved the problems described in Figs. 15 and 16.
You can find the paper at this link: http://dx.doi.org/10.1117/1.3574097.