A model of speed tuning in MT neurons.

We have shown previously that neurons in the middle temporal (MT) area of primate cortex have inseparable spatiotemporal receptive fields-their response profiles exhibit a ridge that is oriented in the spatiotemporal frequency domain, and this orientation predicts the neurons' preferred speed. When measured in spatiotemporal frequency space, such MT spectral receptive field (SRF) properties are closely matched to the spectrum generated by a moving edge. In contrast, V1 neurons have SRF properties that are poorly matched to moving edge spectra, indicating that V1 neurons are not tuned to a particular image speed but rather to specific spatial and temporal frequencies. Here we describe a neural mechanism based directly on the properties of V1 neurons that is able to explain the SRF change that occurs between V1 and MT. We outline the theory behind this transformation and posit an explanation for how the visual system extracts true speed (independent of spatial frequency) from retinal image motion. We tested this speed model against our MT neuron data and found that it provides an excellent account of speed tuning in MT.