Norberto M. Grzywacz

A Model for the Estimate of Local Image Velocity by Cells in the Visual Cortex

Some computational theories of motion perception assume that the first stage en route to this perception is the local estimate of image velocity. However, this assumption is not supported by data from the primary visual cortex. Its motion sensitive cells are not selective to velocity, but rather are directionally selective and tuned to spatio-temporal frequencies. Accordingly, physiologically based theories start with filters selective to oriented spatio-temporal frequencies. This paper shows that computational and physiological theories do not necessarily conflict, because such filters may, as a population, compute velocity locally. To prove this point, we show how to combine the outputs of a class of frequency tuned filters to detect local image velocity. Furthermore, we show that the combination of filters may simulate ‘Pattern’ cells in the middle temporal area (MT), whereas each filter simulates primary visual cortex cells. These simulations include three properties of the primary cortex. First, the spatio-temporal frequency tuning curves of the individual filters display approximate space-time separability. Secondly, their direction-of-motion tuning curves depend on the distribution of orientations of the components of the Fourier decomposition and speed of the stimulus. Thirdly, the filters show facilitation and suppression for responses to apparent motions in the preferred and null directions, respectively. It is suggested that the MT’s role is not to solve the aperture problem, but to estimate velocities from primary cortex information. The spatial integration that accounts for motion coherence may be postponed to a later cortical stage.

Constraints on long range interactions mediating contour detection

Contour detection may be mediated by lateral interactions between neighboring cortical neurons whose receptive fields have collinear axes of preferred orientation. This hypothesis was tested in psychophysical experiments and computer simulations using a contour detection task in which observers searched for groups of Gabor patches that followed spatially extended contour paths embedded in noise consisting of several hundred Gabor patches with random positions and orientations. The orientation-selective units in the simulated neural network were linked by facilitatory interconnections whose strength depended on the geometry (distance, curvature, change in curvature) of smooth curves connecting the orientation axes of units in a pairwise fashion. Psychophysical detection performance was much higher for contour signal groups that followed closed rather than open-ended paths. However, just two sudden changes in orientation of neighboring Gabor patch elements in closed-path contours reduced detection performance to the same levels obtained with open-ended contours. These psychophysical data agreed with the results of the neural network simulations. Furthermore, the simulations also accounted for previous findings that removal of a single Gabor patch element from a closed-path contour group significantly degraded detection performance. We conclude that closure alone is not sufficient to enhance the visibility of a contour. However, if a closed contour meets certain geometric constraints, then lateral interactions based on these constraints can generate facilitation that reverberates around the closed path, thereby enhancing the contours visibility.

Influence of spontaneous activity and visual experience on developing retinal receptive fields

BACKGROUND The role played by early neural activity in shaping retinal functions has not yet been established. In the developing vertebrate retina, ganglion cells fire spontaneous bursts of action potentials before the onset of visual experience. This spontaneous bursting disappears shortly after birth or eye opening. In the present study, we have investigated whether the outgrowth of receptive fields in turtle retinal ganglion cells is affected by early spontaneous bursting or by early visual experience. RESULTS Ganglion cells normally stop bursting spontaneously 2-4 weeks post-hatching, the time when receptive-field areas reach adult size. When turtles are reared in the dark, the spontaneous bursting persists. Concomitantly, receptive-field areas expand to more than twice those observed in normal adults. To test whether chronic blockade of spontaneous bursting inhibits the expansion of developing receptive-field areas, we have exposed the retina to curare, a nicotinic cholinergic antagonist, because spontaneous bursting by ganglion cells requires acetylcholine. Curare was released from Elvax, a slow-release polymer that was implanted in the eye. When spontaneous bursting was chronically blocked with curare in hatchlings, dark-induced expansion of receptive fields was abolished. Moreover, receptive fields of ganglion cells exposed to curare in hatchlings reared in normal light and dark cycles were smaller than normal. CONCLUSIONS These results strongly suggest that early, acetylcholine-dependent spontaneous bursts of activity control the outgrowth of receptive-field areas in retinal ganglion cells. The onset of visual experience induces the disappearance of the immature spontaneous bursts, resulting in the stabilization of receptive-field areas to their mature size.

A winner-take-all mechanism based on presynaptic inhibition feedback

A winner-take-all mechanism is a device that determines the identity and amplitude of its largest input (Feldman and Ballard 1982). Such mechanisms have been proposed for various brain functions. For example, a theory for visual velocity estimate (Grzywacz and Yuille 1989) postulates that a winner-take-all selects the strongest responding cell in the cortexs middle temporal area (MT). This theory proposes a circuitry that links the directionally selective cells in the primary visual cortex to MT cells, making them velocity selective. Generally, several velocity cells would respond, but only the winner would determine the perception. In another theory, a winner-take-all guides the spotlight of attention to the most salient image part (Koch and Ullman 1985). Also, such mechanisms improve the signal-to-noise ratios of VLSI emulations of brain functions (Lazzaro and Mead 1989). Although computer algorithms for winner-take-all mechanisms exist (Feldman and Ballard 1982; Koch and Ullman 1985), good biologically motivated models do not. A candidate for a biological mechanism is lateral (mutual) inhibition (Hartline and Ratliff 1957). In some theoretical mutual-inhibition networks, the inhibition sums linearly to the excitatory inputs and the result is passed through a threshold non linearity (Hadeler 1974). However, these networks work only if the difference between winner and losers is large (Koch and Ullman 1985). We propose an alternative network, in which the output of each element feeds back to inhibit the inputs to other elements. The action of this presynaptic inhibition is nonlinear with a possible biophysical substrate. This paper shows that the new network converges stably to a solution that both relays the winners identity and amplitude and suppresses information on the losers with arbitrary precision. We prove these results mathematically and illustrate the effectiveness of the network and some of its variants by computer simulations.

Temporal coherence theory for the detection and measurement of visual motion

A recent challenge to the completeness of some influential models of local-motion detection has come from experiments in which subjects had to detect a single dot moving along a trajectory amidst noise dots undergoing Brownian motion. We propose and test a new theory of the detection and measurement of visual motion, which can account for these signal-in-Brownian-noise experiments. The theory postulates that the signals from local-motion detectors are made coherent in space and time by a special purpose network, and that this coherence boosts signals of features moving along non-random trajectories over time. Two experiments were performed to estimate parameters and test the theory. These experiments showed that detection is impaired with increasing eccentricity, an effect that varies inversely with step size. They also showed that detection improves over durations extending to at least 600 msec. An implementation of the theory accounts for these psychophysical detection measurements.

Power spectra and distribution of contrasts of natural images from different habitats

Some theories for visual receptive fields postulate that they depend on the image statistics of the natural habitat. Consequently, different habitats may lead to different receptive fields. We thus decided to study how some of the most relevant statistics vary across habitats. In particular, atmospheric and underwater habitats were compared. For these habitats, we looked at two measures of the power spectrum and one of the distributions of contrasts. From power spectra, we analyzed the log-log slope of the fall and the degree of isotropy. From the distribution of contrasts, we analyzed the fall in a semi-log scale. Past studies found that the spatial power spectra of natural atmospheric images fall linearly in logarithmic axes with a slope of about -2 and that their distribution of contrasts shows an approximate linear fall in semi-logarithmic axes. Here, we show that the power spectrum of underwater images have statistically significantly steeper slopes ( approximately -2.5 in log-log axes) than atmospheric images. The vast majority of power spectra are non-isotropic, but their degree of anisotropy is extremely low, especially in atmospheric images. There are also statistical differences across habitats for the distribution of contrasts, with it falling faster for underwater images than for atmospheric ones. We will argue that these differences are due to the optical properties of water and that the differences have relevance for theories of visual receptive fields. These theories would predict larger receptive fields for aquatic animals compared to land animals.

A model of the directional selectivity circuit in retina: transformations by neurons singly and in concert

Publisher Summary This chapter presents a model vertebrate retinal circuit that accounts for the emergence of neurons that selectively respond to stimulus motion in a particular direction in the retina. It discusses three critical elements for directionality of neuron response: an asymmetric input-to-output distribution on the directionally selective neuron dendrite; intracellular resistivity of the dendrite, which allows on-path interactions that depend on stimulus direction; and inhibitory nonlinear shunting of sufficient duration to mask subsequent proximal excitation of the distal tip output. Simulations of morphometrically and biophysically detailed cell models will demonstrate model performance. The chapter also discusses the way this model may work in a developmental context and implications for more general multidimensional filtering within dendritic trees.

Incremental rigidity scheme for recovering structure from motion: position-based versus velocity-based formulations

Perceptual studies suggest that the visual system may use a rigidity assumption in its recovery of three-dimensional structure from motion. Ullman [Perception 13, 255 (1984)] recently proposed a computational scheme that uses this assumption to recover the structure of rigid and nonrigid objects in motion. The scheme assumes the input to be discrete positions of elements in motion, under orthographic projection. We present formulations of Ullmans method that use velocity information and perspective projection in the recovery of structure. Theoretical and computer analyses show that the velocity-based formulations provide a rough estimate of structure quickly but are not robust over an extended time. The stable long-term recovery of structure requires disparate views of moving objects.

Local motion detectors cannot account for the detectability of an extended trajectory in noise

Previous work has shown that a single dot moving in a consistent direction is easily detected among noise dots in Brownian motion (Watamaniuk et al., Vis Res 1995;35:65-77). In this study we calculated the predictions of a commonly-used psychophysical motion model for a motion trajectory in noise. This model assumes local motion energy detectors optimally tuned to the signal, followed by a decision stage that implements the maximum rule. We first show that local motion detectors do indeed explain the detectability of brief trajectories (100 ms) that fall within a single unit, but that they severely underestimate the detectability of extended trajectories that span multiple units. For instance, a 200 ms trajectory is approximately three times more detectable than two isolated 100 ms trajectories presented together within an equivalent temporal interval. This result suggests a nonlinear interaction among local motion units. This interaction is not restricted to linear trajectories because circular trajectories with curvatures larger than 1 degree are almost as detectable as linear trajectories. Our data are consistent with a flexible network that feeds forward excitation among units tuned to similar directions of motion.

Necessity of acetylcholine for retinal directionally selective responses to drifting gratings in rabbit

1 A model for retinal directional selectivity postulates that GABAergic inhibition of responses to motions in the null (anti‐preferred) direction underlies this selectivity. An alternative model postulates that besides this inhibition, there exists an asymmetric, nicotinic acetylcholine (ACh) input from starburst amacrine cells. It is possible for the latter but not the former model that stimuli could exist such that nicotinic blockade eliminates directional selectivity. Such stimuli would drive the cholinergic but not the GABAergic system well. 2 So far, attempts to eliminate directional selectivity with nicotinic blockade have failed, but they always used isolated, moving bars as the stimulus. We confirmed this failure for On‐Off directionally selective (DS) ganglion cells in our preparation of the rabbits retina. 3 However, while recording from these cells, we discovered that nicotinic blockade eliminated directional selectivity to drifting, low spatial frequency sine‐ and square‐wave gratings. 4 This effect was not just due to the smallness of the responses under nicotinic blockade. NMDA blockade caused even smaller responses, but no loss of directional selectivity. 5 This result is consistent with a two‐asymmetric‐pathways model of directional selectivity, but inconsistent with an asymmetric‐GABA‐only model. 6 We conclude that asymmetric nicotinic inputs extend the range of stimuli that can elicit directional selectivity to include moving textures, that is, those with multiple peaks in their spatial luminance profile.