Quantitative Analysis of a Bioplausible Model of Misperception of Slope in the Cafe Wall Illusion

This paper presents a model explaining tilt illusion effect in the Café Wall pattern. In this geometric illusion, we perceive horizontal edges as tilted. We explain this as the result of innate retinal/gangliar visual processing of the pattern. Our bioplausible model is based on a simple early layer using Difference of Gaussian over simple ON-center and OFF-center receptive fields, with a quantification module replacing later layers of a Deep Neural Network. The experimental results show that this bioplausible filtering technique can explain the tilt illusion of the Café Wall pattern. Our statistical analysis of tilt provides a quantitative measurement and an empirically testable prediction for the degree of tilt. This shows that the Difference of Gaussian reveals cues for perception and clues about the illusions we perceive.