Purchase this article with an account.
Tom Putzeys, Robbe Goris, Johan Wagemans, Matthias Bethge; Inferring characteristics of stimulus encoding mechanisms using rippled noise stimuli. Journal of Vision 2009;9(8):349. doi: 10.1167/9.8.349.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Several psychophysical studies have used masking techniques to infer characteristics of stimulus encoding mechanisms underlying early visual processing. These studies typically suggest the existence of multiple frequency- and orientation-selective filters or ‘channels’. To evaluate the usefulness of such a multiple channel encoding front-end in more general models of pattern vision, knowledge about various channel properties is required. Notably, estimates of channel characteristics such as shape and bandwidth vary considerably among studies. One problem is that inferring encoding mechanism characteristics requires (often unwarranted) assumptions regarding various aspects of the visual system (e.g., linearity of contrast processing). Differences in estimates of the channels may reveal important nonlinearities that need to be taken into account. In the present study, we start from reported channel characteristics and traditional assumptions to generate predictions for a new class of stimuli. More specifically, assuming linearity in strong visual noise, the psychophysical channel shape proposed by previous studies can be approximated by a discrete Fourier series. Thus, under the linearity assumption, any given channel shape corresponds to a specific set of Fourier coefficients. A novel kind of stimulus, i.e. rippled noise, is introduced in the present study to estimate these Fourier coefficients. Rippled noise, characterised by a sinusoidally-modulated frequency spectrum, has been used before in hearing research but not in spatial vision. Channel estimates resulting from this new detection-in-noise method will be presented and compared to estimates reported in previous studies.
This PDF is available to Subscribers Only