Temporal noise-reduction filtering of image sequences is commonly appl
ied in medical imaging and other applications, and a common assessment
technique is to measure the reduction in display noise variance. Theo
retically and experimentally, we demonstrate that this is inadequate b
ecause it does not account for the interaction with the human observer
. Using a new forced-choice method we compare detectability of low-con
trast objects and find a noise level for an unfiltered sequence that g
ives the same detectability as the filtered sequence. We report the eq
uivalent detectability noise variance ratio, or EDVR. For a digital lo
w-pass filter that reduces the bandwidth by 1/2, display noise reducti
on predicts an EDVR of 0.5. The measured value averaged over three sub
jects, 0.93+/-0.19, compares favorably with the 0.85 predicted from a
theoretical human observer model, and both are very close to the value
of 1.0 expected for no filtering. Hence, the effective, perceived noi
se is relatively unchanged by temporal low-pass filtering. The computa
tional observer model successfully evaluates a simple low-pass tempora
l filter, and we anticipate that it can be used to predict the observe
r response to other image enhancement filters. (C) 1996 SPIE and IS&T.