Keywords
Visual perception, Biological neural networks, Classical Receptive Field model, Tilt effects, Cafe Wall illusion, Lateral Inhibition of retinal ganglion cells, Difference of Gaussians
Abstract
This abstract explores the tilt effect in a family of Café Wall illusions using a Classical Gaussian Receptive Field model (CRF). Our model constructs an intermediate representation called edge map at multiple scales (Fig. 1) that reveals tilt cues and clues involved in the illusory perception of the Café Wall pattern. We investigate a wide range of parameters of the stimulus including mortar width, luminance, tiles contrast, and phase of the tile displacement (the stimuli in Fig. 2). We show that this simple bioplausible model, simulating the contrast sensitivity of the retinal ganglion cells, can not only detect the tilts in this pattern but also allows us to predict the strength of the illusion and quantify the degree of tilt. We embed the DoG model in a processing pipeline using Hough space for quantitative measurement of tilt angles in the edge map to compare them with the tilt perceived by a human observer. A systematic approach was designed as a classifier (below diagram) for the tilt effect that is described based on the horizontal mean tilts detected at fine scale close to the mortar size, that reflect the ‘foveal tilt effect’ (FTE) as well as the ‘persistence of mortar cues’ (PMC) across multiple scales of the edge maps. Even though the results of our simulations are consistent with previous psychophysical findings reported in the literature across the full range of Café Wall variations tested, it is a priority in our future study to run psychophysical experiments to confirm the precise quantitative predictions made by our model.
Start Date
15-5-2019 11:00 AM
End Date
15-5-2019 11:30 AM
Location
Adelaide, South Australia
Included in
Biological Engineering Commons, Other Computer Engineering Commons, Robotics Commons, Vision Science Commons
Quantified Measurement of the Tilt Effect in a Family of Café Wall Illusions
Adelaide, South Australia
This abstract explores the tilt effect in a family of Café Wall illusions using a Classical Gaussian Receptive Field model (CRF). Our model constructs an intermediate representation called edge map at multiple scales (Fig. 1) that reveals tilt cues and clues involved in the illusory perception of the Café Wall pattern. We investigate a wide range of parameters of the stimulus including mortar width, luminance, tiles contrast, and phase of the tile displacement (the stimuli in Fig. 2). We show that this simple bioplausible model, simulating the contrast sensitivity of the retinal ganglion cells, can not only detect the tilts in this pattern but also allows us to predict the strength of the illusion and quantify the degree of tilt. We embed the DoG model in a processing pipeline using Hough space for quantitative measurement of tilt angles in the edge map to compare them with the tilt perceived by a human observer. A systematic approach was designed as a classifier (below diagram) for the tilt effect that is described based on the horizontal mean tilts detected at fine scale close to the mortar size, that reflect the ‘foveal tilt effect’ (FTE) as well as the ‘persistence of mortar cues’ (PMC) across multiple scales of the edge maps. Even though the results of our simulations are consistent with previous psychophysical findings reported in the literature across the full range of Café Wall variations tested, it is a priority in our future study to run psychophysical experiments to confirm the precise quantitative predictions made by our model.