ABSTRACT
In Chapter 4, a quality loss function was defined to be the relationship between the quality change in JNDs arising from an attribute and the value of an objective metric correlated with that attribute. In that chapter, the quality loss function of a contrast metric was determined, thereby quantifying the loss of quality caused when the preferential attribute of contrast deviated by different amounts from its optimum value for a particular scene and observer. The quality loss function of the contrast metric was fit with a quadratic function primarily for mathematical convenience, and although the psychometric data was fit well (Fig. 4.4), that data spanned a fairly narrow range of quality (ca. 9 JNDs) and so represented a relatively undemanding case. Researchers typically fit each new set of psychometric ratings versus objective metric values with whatever mathematical function seems to be suggested by the data. In some cases, less reliable data have been fit with complex functions to account for saturation and range effects, which were largely artifacts of the psychometric methods employed rather than perceptual phenomena of primary interest. The wide range of properties of the ratings being analyzed has contributed to the proliferation of specialized treatments.
