Abstract
Multiscale spatial filtering models account for a variety of phenomena in human lightness perception, such as Simultaneous Contrast and White's illusion, by combining outputs of oriented filters at multiple scales. However, it was shown that these models were unable to account for spatial frequency specific effects of narrowband noise on White's illusion (Betz, Shapley, Wichmann, & Maertens, 2015). This data showed that noise between 1-5 cpd specifically interfered with White's illusion. The effect did not scale proportionally with stimulus size. To us this suggests a crucial role of luminance edges in lightness perception as proposed by edge-integration models. In these models, lightness is computed by integrating contrast signals at luminance discontinuities (=edges) across space. Here we propose an edge-sensitive mechanism in which edge signals emerge from multiscale filter responses to an actively-sampled, time-varying input signal such as it would result from fixational eye movements (FEMs). Instead of extracting edges explicitly from edge-detectors, our computational approach produces edge signals as a byproduct of using unoriented multiscale filters on these time-varying inputs. We show in simulations that edge signals actively derived from FEMs are most strongly affected by noise between 1-5 cpd. This is consistent with the data of Betz et al. (2015). Furthermore, we find that a broader range of filter scales is sensitive to edge information when using this approach as compared to extracting edges explicitly from edge detectors. We take this as initial evidence that incorporating FEMs in multiscale spatial filtering models may be a fruitful way to model edge sensitivity in human lightness perception. Our results support the proposal that FEMs are part of a dynamic strategy of the visual system to actively encode visual information in space and time.