Understanding image quality

Summary form only given. Image quality research traditionally focuses on the subjective measurement, prediction, or improvement of image quality. The fundamental question of what image quality is, however, has been given surprisingly little attention. In this paper we present a theoretical framework for understanding image quality that is based on a four-point philosophy. The basic definition of image quality that we give is formulated in terms of the degree to which two imposed requirements are satisfied: usefulness (that is, discriminability of image content) and naturalness (identifiability of image content). When an image is considered to be input to the interaction process, these two requirements must be imposed on the image to guarantee the success of the interaction process. In this view, the degree to which an image satisfies the usefulness and naturalness requirements determines the quality of this image. The validity of the above theoretical concept is demonstrated for the case of color reproductions of natural scenes. First, it is shown how the concept is able to explain small but systematic deviations between naturalness and quality judgments for manipulated color reproductions. Second, it is shown how computational algorithms for estimating discriminability and identifiability of image content are able to predict these judgments.