Absolute limits on image processing
David G. Brown, Robert F. Wagner, Mary Pastel Anderson · 1980
The metrology of image processing for medical diagnostic imaging systems can be developed from two different approaches. One method involves the comparison of actual clinical efficacy of one processing scheme versus another. A paradigm for this approach is the ROC analysis technique outlined by Metz and by Swets. The second method, taking as its starting point axioms of information theory proposed by Shannon, calculates the performance of an ideal observer from the system imaging parameters for some idealized imaging task. Image processing can only improve the performance of the human observer up to that of the ideal observer, in the limit of "ideal" processing.