Hh. Barrett et al., OBJECTIVE ASSESSMENT OF IMAGE QUALITY - III - ROC METRICS, IDEAL OBSERVERS, AND LIKELIHOOD-GENERATING FUNCTIONS, Journal of the Optical Society of America. A, Optics, image science,and vision., 15(6), 1998, pp. 1520-1535
We continue the theme of previous papers [J. Opt. Sec. Am. A 7, 1266 (
1990); 12, 834 (1995)] on objective (task-based) assessment of image q
uality. We concentrate on signal-detection tasks and figures of merit
related to the ROC (receiver operating characteristic) curve. Many dif
ferent expressions for the area under an ROC curve (AUC) are derived f
or an arbitrary discriminant function, with different assumptions on w
hat information about the discriminant function is available. In parti
cular, it is shown that AUC can be expressed by a principal-value inte
gral that involves the characteristic functions of the discriminant. T
hen the discussion is specialized to the ideal observer, defined as on
e who uses the likelihood ratio (or some monotonic transformation of i
t, such as its logarithm) as the discriminant function. The properties
of the ideal observer are examined from first principles. Several str
ong constraints on the moments of the likelihood ratio or the log like
lihood are derived, and it is shown that the probability density funct
ions for these test statistics are intimately related. In particular,
some surprising results are presented for the case in which the log li
kelihood is normally distributed under one hypothesis. To unify these
considerations, a new quantity called the likelihood-generating functi
on is defined. It is shown that all moments of both the likelihood and
the log likelihood under both hypotheses can be derived from this one
function. Moreover, the AUC can be expressed, to an excellent approxi
mation, in terms of the likelihood-generating function evaluated at th
e origin. This expression is the leading term in an asymptotic expansi
on of the AUG; it is exact whenever the likelihood-generating function
behaves linearly near the origin. It is also shown that the likelihoo
d-generating function at the origin sets a lower bound on the AUC in a
ll cases. (C) 1998 Optical Society of America.