Digital image processing: effect on detectability of simulated low-contrast radiographic patterns.

Abstract

Detection studies of simulated low-contrast radiographic patterns were performed with a high-quality digital image processing system. The original images, prepared with conventional screen-film systems, were processed digitally to enhance contrast by a windowing technique. The detectability of simulated patterns was quantified in terms of the results of observer performance experiments by using the multiple-alternative forced-choice method. The processed images demonstrated a significant increase in observer detection performance over that for the original images. These results are related to the displayed and perceived signal-to-noise ratios derived from signal detection theory. The improvement in detectability is ascribed to a reduction in the relative magnitude of the human observer''s internal noise after image processing. The measured dependence of threshold signal contrast on object size and noise level is accounted for by a statistical decision theory model that includes internal noise.