Space-Time Receptive Fields of Visual Neurons
by Izumi Ohzawa, email@example.com
indicates MPEG animation file.
indicates Acrobat (PDF) slide file.
indicates PostScript (PS) slide file.
For viewing PDF files, free Acrobat Readers are available from the
Adobe Systems Web site.
Modern Views and Measurements of Visual Receptive Fields
Traditionally, the receptive field of a neuron is defined as the area of
visual field within which visual stimulation influences neural responses.
This classical notion no longer provides an adequate framework for understanding
visual receptive fields. We must consider an additional dimension of time, and
define receptive fields in the joint domain of space and time. For many cells
in the visual cortex, there is no such thing as a unique spatial receptive field.
Slides, animations here attempt to demonstrate this fact in an intuitively
For a written introductory material on this subject,
please read our review paper (on-line version available):
Receptive-field dynamics in the central visual pathways (TINS 1995)
by DeAngelis, Ohzawa, and Freeman.
Overview and Summary
LGN Cells (lateral geniculate nucleus)
Simple Cells (primary visual cortex)
Complex Cells (primary visual cortex)
Space-time separable receptive field of a simple cell
These RFs are expressed as the product of two 1-D functions:
a function of space and a function of time.
Space-time inseparable receptive field of a simple cell
These RFs are oriented and inseparable in the space-time domain. That is, the space-time RF
cannot be expressed as the product of two 1-D functions, one of space and another
of time. The space-time orientation is related to direction selectivity of these simple cells.
Space-time RF of a simple cell in primate V1
This RF animation on Dario Ringach's Web page at
now UCLA shows a
space-time RF obtained from a simple cell in V1. He uses
a novel reverse correlation mapping techinque in the spatial frequency and
orientation domain (using a bunch of sinusoidal gratings of different spatial frequencies
and orientations). The spatial RF profile is reconstructed from these frequency
domain data. In addition to allowing RF reconstructions,
the dynamics of orientation tuning
and spatial frequency tuning may be examined directly
by this technique
of orientation tuning curves by Ringach].
Note that the reconstruction of RFs from the frequency domain data
requires a linearity assumption.
Binocularity and Stereopsis (primary visual cortex)
There is no apparent elongated, oriented subfield structure
for these cells. The RF of complex cells appear to be a broadly
selective Gaussian in two dimensions of space (x, y) when mapped with
a single bar-shaped stimulus.
Binocular receptive fields and stereopsis
Most neurons in the visual cortex are binocular.
We consider receptive field properties with respect to
binocular vision and stereopsis. Inhererently, a complete
description of a binocular RF must be given in the 4-dimensional
space (X, Y, Z, T), where (X, Y) defines the viewer's visual direction,
Z is the distance from the viewer, and T is the time.