Abstract

I will briefly introduce the idea of how we might extract motion using spatio-temporal energy. Spatio-temporal energy models of motion involve breaking down moving patterns into simpler sinusoidal patterns that have their own motion, and it is the spatio-temporal energy of these components that determine the response of V1 neurons. This is an efficient method of encoding pattern motion because for moving sinusoidal images, all measurements of the same velocity lie along a line in spatial frequency/temporal frequency space. I have been interested in how the spatio-temporal energy of these simpler patterns might be combined to predict perceived pattern motion direction. I will describe the two main combination rules: the ‘intersection of constraints’ rule (IOC) and the ‘vector average’ (VA). I will provide empirical evidence showing that the results supporting the vector average do not generalise. In addition, David Alais and myself have shown that both solutions can be simultaneously encoded. This is curious given that they are thought to be mutually exclusive hypotheses. When perceived pattern motion is predicted by the vector average it may be caused by underlying mechanisms that respond to either spatio-temporal energy caused by distortion products, or possibly by my implementation of the intersection of constraints rule that can also encode information corresponding to distortion products. This may explain why vector averaging is not a general property and why both solutions ‘appear’ to be encoded.