Abstract

The coding of stimulus speed has been accepted to be an important component of motion processing in the visual system. How this coding is implemented, however, is still not entirely clear. In terms of its relationship with spatial and temporal contents of moving targets, speed processing is often modeled as a two-stage process (e.g. Heeger, 1990; Smith & Edgar, 1994). In the first stage, initial speed estimates are generated within individual spatial- and temporal-frequency-tuned mechanism. In the second stage, the outputs of multiple spatio-temporal frequency mechanisms from the first stage are combined to produce the speed codes that are then invariant with respect to spatial frequency. This descriptive model for speed coding makes sense from both functional and experimental perspectives. Functionally, speed coding in the visual system should not vary with respect to spatial frequency; otherwise, non-identical speed coding among spatial frequency mechanisms would perceptually make the different spatial frequency components of a single target appear to move incoherently. Experimentally, it has been demonstrated that whereas the responses of neurons at early stages, such as in VI, are spatial-frequency-tuned, the responses of some neurons at later stages, such as in MT, have broader spatial frequency bandwidths (Newsome, Gizzi & Movshon, 1983), which presumably represent the combination of inputs from several lower-level spatial frequency mechanisms.

© 1996 Optical Society of America