Human observer modeling of quantitative image quality in temporally filtered image sequences [fluoroscopy application

Abstract

Temporal filtering is a common method of improving image quality in noisy image sequences, and results are frequently assessed by measuring the reduction in display noise pixel variance. The authors show that this assessment technique is naive because it ignores the interaction of the human visual system with the displayed sequence. Using forced-choice perception experiments, the authors quantitate the effect of three temporal filters on image quality by measuring their effect on detectability of low-contrast stationary disks in noisy image sequences. Filters studied are a pair of symmetric low-pass and high-pass Butterworth filters, and a low-pass first-order recursive filter. The authors compare results to predictions from a human observer model that incorporates a psychophysically-measured spatiotemporal visual system response, as well as to predictions from display noise reduction alone. Measurements agree much more favorably with predictions from the observer model. It is concluded that it is essential to include the human visual response in the evaluation of filtering of noisy image sequences.