Consistency of classification images across noise dimension P Neri |
|---|
University of Aberdeen, United Kingdom
|
| When measuring the tuning characteristics of visual mechanisms using stimulus noise, it is often assumed that a given mechanism will retain stable behaviour regardless of the dimension probed by the applied noise process. For example, the tuning properties of an edge detector should be similar whether we probe its spatial preference using pixel noise, or whether we probe its orientation preference using orientation noise. I will discuss data where this and related predictions are put to direct experimental test by deriving perceptual filters (classification images) using different noise probes. The data demonstrate that some properties of the detection mechanism are stable under different noise manipulations, while others are not. I will then discuss computational models that may offer an explanation for the observed similarities/differences. |
| Up Home |
|---|