Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in

BCCN Sonderseminar

Thursday, 12.07.2007 09 c.t.

Motion processing in Fourier space: Computing local velocities by constructive interference of global signals

by Dr. Babette Dellen
from BCCN Göttingen

Contact person: Theo Geisel

Location

Seminarraum Haus 2, 4. Stock (Bunsenstr.)

Abstract

The large receptive fields of motionsensitive neurons belonging to the avian tectorotundal pathway suggest that global Fourierrelated transformations play an important role in the computation of visual motion. But a problem arises. Even though highly accurate velocity (frequency) measurements can be performed in a global Fourier space, the velocities of the measured signals (Fourier components) cannot be assigned to a single spatial position due to their global character. We suggest that this problem can be resolved by utilizing the principle of constructive interference. In the proposed algorithm, local image velocities are derived from the global Fourier components of the image sequence by searching for those velocities for which the superposition of the corresponding Fourier components leads to constructive interference at the image point. Since the localization of velocities within the image space is derived from global signals, the technique does not suffer the aperture problem and is thus capable of capturing motion signals that take place on different spatiotemporal scales. The algorithm is tested on real image sequences, and the results are compared to stateoftheart techniques for opticflow computation. Plausible neuronal implementations of the algorithm are discussed.

back to overview