Perceptual image difference metrics - a unifying approach to image representation and reproduction

One of the major goals for a wide range of image representation and reproduction techniques is to keep an image processed by an image processing chain perceptually as close as possible to the original. Perceptual differences between two representations or reproductions of an image can be determined by psychophysical experiments or by algorithmic metrics. While being better at capturing perceptual differences, the former are inherently time consuming, non quantitative, and error prone, due to observer subj ectivity. The latter are objective, quantifiable and easy to evaluate numerically, which makes them better suited for performance assessment of image processing algorithms; they are, however, still immature. Therefore, this project has two main goals: the first is to develop new perceptual image difference metrics, and the second is to use them in new ways to improve the state of the art in image representation and reproduction techniques. Hence, in the first part of the project we will focus on the metri cs themselves, by analysing and evaluating current image difference metrics, and by developing new metrics based on improved colour difference metrics, perceptual predictors, saliency maps, and multilevel methods. We will create a good test dataset; this is of paramount importance to the success of the development of new metrics. In the second part of the project, we will use the proposed image difference metrics with numerical optimisation, variational techniques and heuristic search methods to develop n ew frameworks for optimising different image representation and reproduction algorithms. Specifically, these frameworks will be used to optimise algorithms for image compression, high dynamic range tone mapping, colour gamut mapping, and halftoning. The p roject will educate one PhD candidate and three master students, employ one postdoctoral researcher, and facilitate international exchange.