Multi-dimensional optimization of visual content

In order to give consumers the ability to adapt a display to their own image preference, current displays are equipped with a number of knobs, by which the viewer can tune various image quality attributes to his or her preference. Despite of this possibility, many people seem to avoid adjustments, because they do not know how to end up with a satisfying result. Results of two studies on preferred television settings, which we
conducted in our laboratory in order to improve our products, clearly illustrate the difficulties many people encounter when optimizing image preference in a multidimensional parameter space spanned by various display settings. The aim of the first study was to define a number of user profiles that might serve as predefined settings in future TV sets. Therefore, we investigated how a large group of users adjusted brightness, contrast, sharpness, saturation and white point of a
conventional LCD-TV. In total, 176 participants started from a very poor image quality and were asked to adjust the TV to their own preference. Most people were unable to do so, and ended up with an image that was not pleasing them. In a second part of the study, the task was simplified: another 200 participants started from an acceptable image quality and had to further refine it by means of a series of one-dimensional
optimizations. In this way, substantially more people were able to optimize the image to
a satisfying result.
In the second study, participants were asked to adjust gain and offset of a Philips nineview lenticular auto-stereoscopic 3D display to their preference. The gain setting
affected the total range of stereoscopic depth that could be displayed, whereas the offset
setting affected what part of the depth range was displayed in front or behind the screen. At first, 24 participants were requested to tune both gain and offset simultaneously. This resulted for many participants in inconsistent results. Therefore, the task was simplified. Another 21 participants were asked to tune the gain for some fixed values of the offset, and the offset for some fixed values of the gain. This largely reduced the spread in the data. The results of both studies clearly show that in general people are not able to optimize display settings when navigating in a multi-dimensional space. It seems that the simultaneous optimisation of even as few as two dimensions is already too difficult. This
may be a consequence of (1) the limited ability of humans to acquire a correct mental map of the multi-dimensional space, or of (2) limitations in the visual memory, though which viewers are unable to accurately recall the quality of the previous image in the adjustment process. As a result, participants tend towards the middle of a scale, even if this position on the scale is not an acceptable setting.