Table of Contents

Decoding the Reverse Phi Illusion

After diving into the code for reverse phi motion, I found myself pondering its intricacies for days. While I can articulate the phenomenon in terms of programming or mathematics, the scientific literature seemed to offer limited insights into whether this illusion is purely retinal or a result of complex neural interactions, such as those involving the geniculate nucleus or the motion-sensitive neurons in V1.

an example of Reverse Phi illusion I coded in Rust Flynn & Shapiro, 2018

A Glimpse into Comparative Vision Science

While exploring this topic, I stumbled upon an intriguing article titled "Common Circuit Design in Fly and Mammalian Motion Vision". Although the paper primarily focuses on flies and mice, its findings can be extrapolated to human vision.

ON and OFF Pathways: The Universal Language of Vision

Both flies and humans possess separate pathways for processing ON and OFF signals generated by photoreceptors that respond to changes in luminance. These pathways originate at different synapses in the visual system and have distinct anatomical and functional properties. According to a 2017 study, these ON and OFF pathways are sensitive to specific interactions, which may also be applicable to human psychophysics.

Demo Example

To better understand ON and OFF pathways, check out this interactive shader: ShaderToy Example. This shader illustrates how ON pathways are more activated by a bright stimulus on a dark background, while OFF pathways are more activated by a dark stimulus on a bright background.

Note: That shader code not belong to me!

The Role of Correlation-Based Algorithms

The article also discusses how both species utilize correlation-based algorithms for motion detection. These algorithms compare temporal changes in luminance at different spatial locations using a multiplication or correlation process. This is in line with another study that showed equal efficiency in correlating dots of opposite contrast and of similar contrast in reverse-phi motion stimuli.

Reverse Phi: More Than Just Stimulus Properties

In essence, reverse phi motion is not merely a result of low-level stimulus properties. It appears to be a complex interplay between ON and OFF signals processed through discrete pathways. These interactions can even reverse the direction of motion when contrast changes accompany the discrete motion, as suggested by biophysical simulations.

Why Parallel Processing?

You might wonder why our visual system even employs this kind of parallel processing. The authors of the article provide an insightful explanation. They suggest that separating ON and OFF signals simplifies the task for motion-sensitive neurons, making it easier to correlate two positive input signals. This seems to resolve a significant biophysical challenge in implementing correlation mechanisms.

Final Thoughts: The Elegance of ON and OFF Pathways in Motion Perception

As we draw this exploration to a close, let's revisit a profound insight from the paper "Common Circuit Design in Fly and Mammalian Motion Vision". The paper asks an essential question: What could be the advantage of separating ON and OFF signals in both the mouse and fly visual systems?

The separation of ON and OFF pathways in our visual system serves a practical, computational purpose. Motion perception, at its core, involves the temporal correlation of similar events at distinct spatial locations. When a bright object moves across our field of vision, the neurons responsible for detecting this motion have to process both the increasing and the decreasing luminance as the object passes by.

The challenge lies in the biophysics: there's no known mechanism that allows a postsynaptic neuron to get excited both when two inputs increase and when they decrease their membrane potential. This is where the genius of biological design comes into play. By segregating ON and OFF signals, each dealing exclusively with brightness increments or decrements, the visual system effectively simplifies the task for motion-sensitive neurons. Now, they only have to correlate two positive input signals, significantly easing the computational burden.