Selecting significant colors from a complex image for image quality modeling

Judgments of complex images differ from those of uniform color samples in several important respects. One such difference is that a complex image is formed of a large number of discrete color elements. Observer judgments are based not on assessment of each discrete element but of a much smaller number of salient features. The judgment process can be considered as the selection of such features followed by the judgment of particular quality attributes for these features. Modeling the judgment process thus requires a set of well-defined quality attributes together with a methodology for the selection of salient features and their relative importance. In this project, a method of selecting colors within an image was considered. A number of measurement locations within a complex image were selected, and the color of these locations was measured on a series of reproductions. The reproductions were judged by a panel of expert observers for their match to a proof, using a category scaling of several image quality attributes. By comparing the measured color differences with the visual judgments it was possible to determine which locations carried the greatest weight in the assessments. It was also possible to make a limited prediction of the visual judgments from the measurements.