MEO_Gabor

iModel

Supported by Wellcome Trust

MEO_Gabor
Opponent Motion Energy, Gabor Filter

Summary

This opponent motion energy (ME) model is perhaps the most idealized model of cortical direction selectivity. It was introduced by Adelson and Bergen (1985) and differs from their physiologically plausible ME model because its linear filters are 3-D Gabor functions (Gaussian times sinusoid). This model belongs to the class known as elaborated Reichardt detectors (van Santen and Sperling, 1984; Adelson and Bergen, 1985; van Santen and Sperling, 1985). Essentially, the model consists of the difference between the outputs (before spike generation) of two ME_Gabor (i.e., non-opponent) models that have opposite preferred directions. Like the ME_Gabor model, it provides an inadequate account of typical DS cells in macaque V1 and V5/MT (e.g., Bair and Movshon, 2004).

Results

Please get a Java compatible browswer to see this.

References

Adelson EH, Bergen JR (1985) Spatiotemporal energy models for the perception of motion. J Opt Soc Am A 2:284-299.
Bair W, Movshon JA (2004) Adaptive temporal integration of motion in direction-selective cells in macaque visual cortex. J Neurosci 24:7305--7323.
Grzywacz NM, Yuille AL (1990) A model for the estimate of local image velocity by cells in the visual cortex. Proc Roy Soc Lond B 239:129-161.
van Santen JPH, Sperling G (1984) Temporal covariance model of human motion perception. J Opt Soc Am A 1:451--473.
van Santen JPH, Sperling G (1985) Elaborated Reichardt detectors. J Opt Soc Am A 2:300--321.

The visual stimulus is processed (convolved) by four linear Gabor filters (icons show x-t slices of 3D filters) to yield the signals fpe (filter preferred even), fpo (filter preferred odd), fae (anti-preferred even) and fao (anti-preferred odd). Each filter output is squared, and the signals are added for each quadrature pair to give the motion energy in the preferred direction, mep, and the anti-preferred direction, mea. These two oppositely tuned ME signals are subtracted to give the opponent motion energy, meopp.

The opponent signal (meopp) is then offset, scaled and half-wave rectified, and it is used to drive a Poisson spiking mechanism. The spikes are time shifted to simulate a neurobiological latency. See the model (.moo) files for the parameters that govern these computations.