Purchase this article with an account.
Stanley Klein, Thom Carney, Hairan Zhu, Sean Bupara; Using Repeated Noise to Look Inside the Box. Journal of Vision 2012;12(9):317. doi: 10.1167/12.9.317.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The use of noise has been a powerful tool in learning about nonlinear stages of visual processing. We have extended the approach of Petrov, Dosher & Lu (2005, 2006) to enable us to gather more accurate data and extend their analysis in a manner that enables us to isolate stages of processing. Petrov measured Gabor orientation discrimination in periphery with added oriented noise. Observers reported target Gabor orientation (left/right) for several target contrasts in the presence of oriented noise background that was congruent or incongruent with the target orientation. In our experiments the following modifications were made: noise orientation never shifted, we included zero contrast Gabor stimuli for classification images, confidence ratings as well as orientation ratings were given, the same noise was typically used for the two locations and two target orientations and across different runs. Otherwise in a single run different noise was used. The purpose of using the same noise repeatedly in randomized order was to gain accurate responses with small variability.
The repeated noise across locations, orientations and runs enabled d' and bias to be calculated for each noise sample, giving the following results and conclusions: 1)One surprising finding was that d' and bias were very different in the two peripheral locations in 4 of the 5 subjects. 2)The orientation/confidence ratings reveal nonlinearities and a jaggedness cue that eliminated the simple model proposed by Petrov et al. 3)Repeated noise across runs revealed factors that changed during perceptual learning, the original goal of the study. 4)Decision stage weights could be shown to be different for different noise samples, a powerful addition for studies using noise. Another difference from the original Petrov et al. results was that by our reaching asymptotic behavior after learning, the filter weights for the congruent orientation were much weaker than for the incongruent orientation.
Meeting abstract presented at VSS 2012
This PDF is available to Subscribers Only