Greetings!
I am conducting applied research on the efficacy of texture-(mis)matching as a
means of reducing (or increasing) the "conspicuity" (d', detectibility,
discriminability) of an image subregion (either intrinsically or with respect to
its background surround). The idea is to use one or more computational metrics
of texture (e.g., greylevel co-occurrences, Markov random fields, etc) to
characterize the surround, then suitably modify the image subregion to match the
surround texture metric(s). This procedure implicitly affects the
discriminability measure (i.e., comparison of subregion and surround d') rather
than the instrinsic d'.
Alternately, I could utilize a computational (early) vision model's measure of
d' in the subregion modification process to directly reduce (or increase) its d'
(or its (mis)match to the average surround d') -- at far greater computational
cost.
I have 2 questions regarding the above, and would greatly appreciate any
comments:
Q1. What differences if any would one anticipate to result between these two
general approaches (i.e., texture-matching versus direct d' metric). I can
anticipate that in a fully unconstrained modification process, the results would
be essentially identical for discriminability. I am particularly interested in
the differences occurring when the modification process is constrained. In a
constrained situation, should minimization of instrinsic d' (detectibility) take
priority over minimization of discriminability?
Q2. Can anyone point me to more or less freely available computational early
vision models suitable for this type of computation? I need only deal with
static imagery, although inclusion of color and motion is welcome. I just
downloaded Landy's EVE software (cf: references below), but am presently unaware
of alternatives.
Thanks!!
EVE References:
==============
Landy, M. S. (1988). The EVE Early Vision Emulation Software: Reference Manual.
Mathematical Studies in Perception and Cognition 88-11, New York University.
Manovich, L. Z., and Landy, M. S. (1988). EVE: Software for psychophysical
modeling.
"Investigative Ophthalmology & Visual Science (supp.)"
Frank Iannarilli
Aerodyne Research, Inc. 45 Manning Rd. Billerica, MA 01821 USA
franki@aerodyne.com