Direction and speed selectivity properties for spatio-temporal receptive fields according to the generalized Gaussian derivative model for visual receptive fields

TL;DR

The study addresses how direction and speed selectivity arise from the geometry of spatio-temporal receptive fields modeled by the generalized Gaussian derivative framework. It derives closed-form direction-speed selectivity for velocity-adapted simple cells across orders of spatial differentiation and analyzes complex cells via quasi-quadrature integration, highlighting how elongation and differentiation order shape selectivity. The results align qualitatively with neurophysiological findings on velocity-tuned neurons in V1/MT and support the Galilean covariance hypothesis for receptive fields. The work provides a theoretical bridge between receptive-field geometry and motion processing, and suggests concrete neurophysiological experiments to quantify these relationships using reconstructed 2+1-D receptive fields.

Abstract

This paper gives an in-depth theoretical analysis of the direction and speed selectivity properties of idealized models of the spatio-temporal receptive fields of simple cells and complex cells, based on the generalized Gaussian derivative model for visual receptive fields. According to this theory, the receptive fields are modelled as velocity-adapted affine Gaussian derivatives for different image velocities and different degrees of elongation. By probing such idealized receptive field models of visual neurons to moving sine waves with different angular frequencies and image velocities, we characterize the computational models to a structurally similar probing method as is used for characterizing the direction and speed selective properties of biological neurons. By comparison to results of neurophysiological measurements of direction and speed selectivity for biological neurons in the primary visual cortex, we find that our theoretical results are consistent with (i) velocity-tuned visual neurons that are sensitive to particular motion directions and speeds, and (ii) different visual neurons having broader vs. sharper direction and speed selective properties. Our theoretical results in combination with results from neurophysiological characterizations of motion-sensitive visual neurons are also consistent with a previously formulated hypothesis that the simple cells in the primary visual cortex ought to be covariant under local Galilean transformations, so as to enable processing of visual stimuli with different motion directions and speeds.

Figures (9)

• Figure 1: Non-causal joint spatio-temporal receptive fields over a 1+1D spatio-temporal domain in terms of the second-order spatial derivative of the form $T_{xx,\text{norm}}(x, t;\; s, \tau, v)$ according to (\ref{['eq-spat-temp-RF-model-der-norm-caus']}) of the product of a velocity-adapted 1-D Gaussian kernel over the spatial domain and the non-causal temporal kernel over the temporal domain according to (\ref{['eq-non-caus-temp-gauss']}). The spatio-temporal receptive fields are shown for different values of the spatial scale parameter $\sigma_x = \sqrt{s}$ and the temporal scale parameter $\sigma_t = \sqrt{\tau}$ in dimensions of $[\hbox{length}]$ and $[\hbox{time}]$. (Horizontal axes: Spatial image coordinate $x \in [-10, 10]$. Vertical axes: Temporal variable $t \in [-4, 4]$.)
• Figure 3: Schematic illustration of the modelling situation studied in the theoretical analysis, where the coordinate system is aligned to the preferred orientation $\varphi = 0$ of the receptive field, and the receptive field is then exposed to a moving sine wave pattern with inclination angle $\theta$. In this figure, the sine wave pattern is schematically illustrated by a set of level lines, overlayed onto a few level curves of a first-order affine Gaussian derivative kernel. (Horizontal axis: spatial coordinate $x_1$. Vertical axis: spatial coordinate $x_2$.)
• Figure 4: Graphs of the direction selectivity for velocity-tuned models of simple cells based on (left column) first-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, (middle left column) second-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, (middle right column) third-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, or (right column) fourth-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, shown for different values of the degree of elongation $\kappa$ between the spatial scale parameters in the vertical vs. the horizontal directions. Observe how the direction selectivity varies strongly depending on the eccentricity $\epsilon = 1/\kappa$ of the receptive fields. (top row) Results for $\kappa = 1$. (second row) Results for $\kappa = 2$. (third row) Results for $\kappa = 4$. (bottom row) Results for $\kappa = 8$.
• Figure 5: Graphs of the direction selectivity for velocity-tuned models of simple cells based on (left column) first-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, (middle left column) second-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, (middle right column) third-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, or (right column) fourth-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, shown for different values of the ratio $r$ in the relationship $u = r \, v$ between the speed $u$ of the motion stimulus and the speed $v$ of the receptive field, for a fixed value of $\kappa = 2$ between the spatial scale parameters in the vertical vs. the horizontal directions. Observe how the direction selectivity varies strongly depending on the eccentricity $\epsilon = 1/\kappa$ of the receptive fields. (top row) Results for $r = 1/4$. (second row) Results for $r = 1/2$. (third row) Results for $r = 1$. (fourth row) Results for $r = 2$. (bottom row) Results for $r = 4$.
• Figure 6: Graphs of the speed selectivity curves$R_{\varphi^m}(r)$ according to (\ref{['eq-rcurve-order1']}), (\ref{['eq-rcurve-order2']}), (\ref{['eq-rcurve-order3']}) and (\ref{['eq-rcurve-order4']}) for the inclination angle $\theta = 0$ for velocity-tuned models of simple cells based on (left) first-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, (middle left) second-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, (middle right) third-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, or (right) fourth-order directional derivatives of affine Gaussian kernels combined with zero-order temporal Gaussian kernels, shown for different values of the parameter $r$ in the relationship $u = r \, v$ between the speed $u$ of the motion stimulus and the speed $v$ of the receptive field. Notably, these measures are independent of the degree of elongation $\kappa$ of the spatio-temporal receptive fields. (Horizontal axes: relative velocity parameter $r \in [0, 4]$. Vertical axes: velocity sensitivity function $R_{\varphi^m}(r)$.)