Elder J, Morgenstern Y, 2007, "Nonlinear pooling mechanisms underlying edge detection" Perception 36 ECVP Abstract Supplement
Nonlinear pooling mechanisms underlying edge detection
J Elder, Y Morgenstern
We investigated pooling mechanisms underlying detection of luminance edges. Classification images for detection of a vertical edge were estimated over a range of noise contrasts. Estimated integration fields are well-approximated by 2-D Gaussian derivative filters. While the spatial-frequency tuning of these filters is roughly consistent with physiological data, the estimated filters are much longer than receptive fields found in early visual cortex. Are these long integration fields the result of nonlinear pooling over more localized mechanisms? We evaluated a pooling model in which local filter responses were half-wave rectified and combined by Minkowski summation. We found that the nonlinear pooling model is more predictive of the trial-by-trial human data than the standard linear template model. Maximum-likelihood estimates of model parameters indicate length tuning of the local mechanisms ranging from 0.6 to 2.3 deg (Gaussian space constant) with an optimal pooling exponent of roughly 2, consistent with an energy model of spatial integration.
[Supported by NSERC and PREA.]
These web-based abstracts are provided for ease of seaching and access, but certain aspects (such as as mathematics) may not appear in their optimum form. For the final published version of this abstract, please see
ECVP 2007 Abstract Supplement (complete) size: 2091 Kb