Computational models of coherent and transparent plaid motion

Area of Science:

• * Visual neuroscience
• * Computational vision
• * Psychophysics

Background:

• * The motion of superimposed gratings (plaids) can be perceived as either a single coherent movement or as two distinct transparent motions.
• * This perceptual shift depends on factors like relative grating contrast and spatial frequency.
• * Existing computational models for motion transparency have limitations, often requiring complex forms with many unknowns.

Purpose of the Study:

• * To propose and validate a computational model for motion transparency in plaid patterns.
• * To investigate how variations in grating contrast and frequency influence the selection between coherent and transparent motion perception.
• * To explore the neural network implementation of such a model and its implications for visual perception.

Main Methods:

• * Development of a stabilized quadratic computational model for motion transparency, incorporating constraints to detect dual velocities.
• * Realization of the model using a two-layer linear feedforward network.
• * Analysis of model selection criteria when multiple models fit image data without error.

Main Results:

• * The proposed model successfully explains the transition from coherent to transparent motion perception in plaids.
• * Adjustments in relative grating contrast and frequency were shown to be key factors driving this perceptual switch.
• * The feedforward network architecture provides a plausible mechanism for implementing the computational model.

Conclusions:

• * A computational framework, realizable by a feedforward network, can account for plaid motion perception phenomena.
• * The model offers an explanation for why the visual system may favor transparency over coherence under specific stimulus conditions.
• * This work contributes to understanding the computational principles underlying visual motion perception and model selection in the brain.