Hi,
We (colleagues/myself) are attempting to computationally optimize
(minimize) the "pre-attentive" conspicuity of an opaque pattern (blob of
pixels):
1. as (occlusionally) inserted statically against a natural outdoor
background;
2. and also when moving across scene (within a glimpse period).
We wish to ask here for opinions on the proper/useful conspicuity measures.
We employ static conspicuity measures which turn out to be quite
efficacious: blob-to-local-background mismatch in mean, variance, and
2-order texture, for each of three color channels. The texture mismatch
measure is based on a gaussian markov random field (GMRF) parameterization
of local background texture (essentially 2-point correlation over various
lags).
Since we are optimizing, such measures need only be relative, and
monotonically related to conspicuity (i.e., we don't need absolute d'
measures). The scaling (weighting) factors applied to each metric above
are varied empirically to each optimization case (given background).
Our present measure for motion detectibility is:
a. apply a bandpass temporal filter on the motion sequence of images (of
moving blob against fixed background)
b. take absolute value of signed resultant, and pool spatially.
We attempt to minimize this "motion energy". The sequence is typically
several frames spanning a glimpse period of about 100msec. The temporal
impulse response filter is from [Watson & Ahumada (1985)] attempting to
mimick flicker response.
Questions:
1. How sensible is our motion detectibility measure? Any better
suggestions?
2. Is there a "rule of thumb" magnitude relationship between motion and
static measures of conspicuity? IOW, what is the ratio of absolute motion
to absolute static conspicuities?
3. Is it sensible (as typically assumed) that motion cues ensue only from
luminance representations, and not chromatic?
Regards,
Frank J. Iannarilli, franki@aerodyne.com
Aerodyne Research, Inc., 45 Manning Rd., Billerica, MA 01821 USA
www.aerodyne.com/cosr.html
Frank J. Iannarilli, franki@aerodyne.com
Aerodyne Research, Inc., 45 Manning Rd., Billerica, MA 01821 USA
www.aerodyne.com/cosr.html
This archive was generated by hypermail 2b29 : Tue Mar 27 2001 - 15:13:30 PST