ABSTRACT
This chapter addresses some of the generic aspects of testing and revising objective metrics based upon comparison of predictions with quality loss functions determined in calibrated psychometric experiments. The concepts reviewed in this chapter are applied frequently elsewhere in Part II of this book. The examples shown in this chapter are drawn from studies of streaking and banding, two digital artifacts that are distinct types of noise differing from the more common isotropic noise exemplified by film granularity. A comparison of these three kinds of noise is provided in Sect. 13.2. Section 13.3 describes the use of series of levels varying only in the magnitude of the attribute under study, to establish a reference quality loss function along the primary dimension of the experimental design. The utility of comparing data from levels varying in other attribute dimensions with the reference quality loss function is demonstrated in Sect. 13.4. Finally, in Sect. 13.5, an example of the identification of limitations of objective metrics is presented.
