Abstract
Previous work (Kersten, 1987; Taylor et al., 2003) has shown that the summation of spatial frequency information in one-dimensional noise patterns is well described by an ideal observer, suggesting that human observers may sum spatial frequency information optimally over a six-octave wide band. This result is surprising, given that many models of vision contain independent channels that only sum probabilistically (Graham, 1989). However, the previous studies used a restricted set of conditions; specifically, the center frequency was at or near the peak of the CSF. To further test the idea of optimal summation, we had observers perform a detection task with band-pass Gaussian white noise centered at 15 cycles/degree. Spatial frequency bandwidth varied from one-half to four octaves. Stimuli were presented for 200ms in a two interval forced-choice task. Unlike what was found for noises centered at 5 cycles/degree, we found that detection thresholds were not consistent with ideal frequency summation. Hence, the ideal frequency summation reported previously does not generalize to other bands of spatial frequency. Currently we are using the response classification technique to reveal the perceptual template for the detection of these patterns. We will discuss how the frequency summation data can be explained using a standard multiple-channel model (Wilson & Gelb, 1984).
Supported by the Canada Research Chair program and NSERC grants OGP0042133 & OGP0105494