Receptive Fields Of Simple Cells Research Articles

On the role of the scale invariance and the independent components of natural scenes on the receptive fields of simple cells

On the role of the scale invariance and the independent components of natural scenes on the receptive fields of simple cells

Predictive Feedback Can Account for Biphasic Responses in the Lateral Geniculate Nucleus

Biphasic neural response properties, where the optimal stimulus for driving a neural response changes from one stimulus pattern to the opposite stimulus pattern over short periods of time, have been described in several visual areas, including lateral geniculate nucleus (LGN), primary visual cortex (V1), and middle temporal area (MT). We describe a hierarchical model of predictive coding and simulations that capture these temporal variations in neuronal response properties. We focus on the LGN-V1 circuit and find that after training on natural images the model exhibits the brain's LGN-V1 connectivity structure, in which the structure of V1 receptive fields is linked to the spatial alignment and properties of center-surround cells in the LGN. In addition, the spatio-temporal response profile of LGN model neurons is biphasic in structure, resembling the biphasic response structure of neurons in cat LGN. Moreover, the model displays a specific pattern of influence of feedback, where LGN receptive fields that are aligned over a simple cell receptive field zone of the same polarity decrease their responses while neurons of opposite polarity increase their responses with feedback. This phase-reversed pattern of influence was recently observed in neurophysiology. These results corroborate the idea that predictive feedback is a general coding strategy in the brain.

Modeling neuronal response to disparity gradient

There is a rich literature of physiological studies that a subset of neurons in visual cortices is discriminative of 3-D surface orientation using only the disparity gradient information. One of the physiological models to account for this sensibility to surface slant is the dif-frequency disparity model. Although this model is physiologically plausible, no computational analysis is available to explain how first-order-disparity sensitive neurons detect slanted surface. In this paper a computational model based on the dif-frequency disparity model is presented. In particular, analytical expressions that fit well with neuronal responses to broadband stimuli are obtained when simple cell receptive field is described by log-Gabor filters. It is shown with mathematical analysis and numerical simulations that our proposed model can not only account for physiological data of neuronal response to surface slant but also detect disparity gradient from random dot and sinusoidal grating stereograms.

Training image-analysis bases improves ‘image fusion’

Modern technology has enabled development of low-cost wireless imaging sensors of various modalities that can be deployed for monitoring scenes. This advance has been strongly motivated by both military and civilian applications, including those related to health care, battlefield surveillance, and environmental monitoring. Multimodal sensors provide diverse degradation, thermal, and visual characteristics. Image fusion combines visual information from various sources into a single representation to facilitate processing by an operator or computervision system. Fusion techniques can be divided into spatial and transform-domain methods.1 The latter enable efficient identification of an image’s salient features. Transformations that have been suggested for image fusion include dual-tree wavelet transforms and pyramid decomposition.1 We recently proposed2, 3 an image-fusion framework—featuring improved performance—that is based on image-analysis bases trained using independent-component analysis (ICA). Receptive fields of simple cells in the mammalian primary visual cortex are usually spatially localized, oriented, and bandpass. Such filter responses can be derived from unsupervised learning of independent visual features or sparse linear codes for natural scenes.4 Basis training employing ICA4 for image denoising through sparse-code shrinkage improved performance with respect to the use of wavelets. The bases were trained by extracting a population of local patches from similar-content images and then processed by the FastICA algorithm4 to estimate the transformation and its inverse. ICA bases are closely related to wavelets and Gabor functions because they represent localized-edge features. They havemore degrees of freedom than wavelets, however, because they adapt to arbitrary orientations. Discrete and dual-tree wavelet transforms have only two and six distinct orientations, respectively.4 ICA bases do not offer a multilevel representation—as do wavelets or pyramid decomposition—nor are they shift invariant. This invariance can Figure 1. Proposed image-fusion framework. T{ } and T−1{ }: Independent-component analysis (ICA)-trained transformations and their inverse. uk(t): Image coefficients in the ICA domain.

A sparse coding model with imperfect feed-forward circuitry

Event Abstract Back to Event A sparse coding model with imperfect feed-forward circuitry Although simple receptive fields are approximately outlined by a small set of highly specific geniculocortical afferents (Reid and Alonso, 1995, Alonso et al. 2001), the feed-forward wiring is often only an imperfect match of the receptive field, suggesting that neural response properties are also shaped by intracortical processing. While existing sparse coding network models successfully explain the emergence of simple receptive fields, they predict that receptive fields are shaped almost exclusively by feed-forward connections, e.g., (Olshausen and Field, 1996, Rehn and Sommer, 2007). Here we ask if models principled by efficient coding can reproduce response properties from real neurons even though the feedforward weights in the model match the measured receptive fields only imperfectly. We present a novel neural network model for sparse coding based on compressed sensing, a computational principle recently developed in engineering for data compression (Candès and Romberg, 2006). The model exhibits smooth and more realistic simple cell receptive fields than its "imperfect" feed-forward connections would predict, compare the two panel of Fig. 1 of the supplement. Thus, the model demonstrates that efficient coding schemes exist in which feedback connections are critical for shaping the receptive fields. We use simulation experiments with our model to characterize deviations between feed-forward circuitry and measured receptive fields.

Binocular robot vision emulating disparity computation in the primary visual cortex

We designed a VLSI binocular vision system that emulates the disparity computation in the primary visual cortex (V1). The system consists of two silicon retinas, orientation chips, and field programmable gate array (FPGA), mimicking a hierarchical architecture of visual information processing in the disparity energy model. The silicon retinas emulate a Laplacian-Gaussian-like receptive field of the vertebrate retina. The orientation chips generate an orientation-selective receptive field by aggregating multiple pixels of the silicon retina, mimicking the Hubel-Wiesel-type feed-forward model in order to emulate a Gabor-like receptive field of simple cells. The FPGA receives outputs from the orientation chips corresponding to the left and right eyes and calculates the responses of the complex cells based on the disparity energy model. The system can provide the responses of complex cells tuned to five different disparities and a disparity map obtained by comparing these energy outputs. Owing to the combination of spatial filtering by analog parallel circuits and pixel-wise computation by hard-wired digital circuits, the present system can execute the disparity computation in real time using compact hardware.

Internal Spatial Organization of Receptive Fields of Complex Cells in the Early Visual Cortex

The receptive fields of complex cells in the early visual cortex are economically modeled by combining outputs of a quadrature pair of linear filters. For actual complex cells, such a minimal model may be insufficient because many more simple cells are thought to make up a complex cell receptive field. To examine the minimalist model physiologically, we analyzed spatial relationships between the internal structure (subunits) and the overall receptive fields of individual complex cells by a two-stimulus interaction technique. The receptive fields of complex cells are more circular and only slightly larger than their subunits in size. In addition, complex cell subunits occupy spatial extents similar to those of simple cell receptive fields. Therefore in these respects, the minimalist schema is a fair approximation to actual complex cells. However, there are violations against the minimal model. Simple cell receptive fields have significantly fewer subregions than complex cell subunits and, in general, simple cell receptive fields are elongated more horizontally than vertically. This bias is absent in complex cell subunits and receptive fields. Thus simple cells cannot be equated to individual complex cell subunits and spatial pooling of simple cells may occur anisotropically to constitute a complex cell subunit. Moreover, when linear filters for complex cell subunits are examined separately for bright and dark responses, there are significant imbalances and position displacements between them. This suggests that actual complex cell receptive fields are constructed by a richer combination of linear filters than proposed by the minimalist model.

Robustness of neural codes and its implication on natural image processing.

In this study, based on the view of statistical inference, we investigate the robustness of neural codes, i.e., the sensitivity of neural responses to noise, and its implication on the construction of neural coding. We first identify the key factors that influence the sensitivity of neural responses, and find that the overlap between neural receptive fields plays a critical role. We then construct a robust coding scheme, which enforces the neural responses not only to encode external inputs well, but also to have small variability. Based on this scheme, we find that the optimal basis functions for encoding natural images resemble the receptive fields of simple cells in the striate cortex. We also apply this scheme to identify the important features in the representation of face images and Chinese characters.

Standing Waves and Traveling Waves Distinguish Two Circuits in Visual Cortex

The visual cortex represents stimuli through the activity of neuronal populations. We measured the evolution of this activity in space and time by imaging voltage-sensitive dyes in cat area V1. Contrast-reversing stimuli elicit responses that oscillate at twice the stimulus frequency, indicating that signals originate mostly in complex cells. These responses stand clear of the noise, whose amplitude decreases as 1/frequency, and yield high-resolution maps of orientation preference and retinotopy. We first show how these maps are combined to yield the responses to focal, oriented stimuli. We then study the evolution of the oscillating activity in space and time. In the orientation domain, it is a standing wave. In the spatial domain, it is a traveling wave propagating at 0.2-0.5 m/s. These different dynamics indicate a fundamental distinction in the circuits underlying selectivity for position and orientation, two key stimulus attributes.

Melting the Iceberg: Contrast Invariance in Visual Cortex

Neurons in visual cortex maintain their selectivity for stimulus orientation despite wide variations in stimulus contrast. Achieving this invariance is a challenge because of the iceberg effect: subthreshold responses are broader in selectivity than firing responses, and increasing contrast would bring them above threshold. An article in this issue of Neuron by Finn et al. explains how neurons solve this problem using simple mechanisms: contrast-gain control, additive noise, and firing threshold.

The Emergence of Contrast-Invariant Orientation Tuning in Simple Cells of Cat Visual Cortex

Simple cells in primary visual cortex exhibit contrast-invariant orientation tuning, in seeming contradiction to feed-forward models that rely on lateral geniculate nucleus (LGN) input alone. Contrast invariance has therefore been thought to depend on the presence of intracortical lateral inhibition. In vivo intracellular recordings instead suggest that contrast invariance can be explained by three properties of the excitatory pathway. (1) Depolarizations evoked by orthogonal stimuli are determined by the amount of excitation a cell receives from the LGN, relative to the excitation it receives from other cortical cells. (2) Depolarizations evoked by preferred stimuli saturate at lower contrasts than the spike output of LGN relay cells. (3) Visual stimuli evoke contrast-dependent changes in trial-to-trial variability, which lead to contrast-dependent changes in the relationship between membrane potential and spike rate. Thus, high-contrast, orthogonally oriented stimuli that evoke significant depolarizations evoke few spikes. Together these mechanisms, without lateral inhibition, can account for contrast-invariant stimulus selectivity.

Face recognition using localized features based on non-negative sparse coding

Neural networks in the visual system may be performing sparse coding of learnt local features that are qualitatively very similar to the receptive fields of simple cells in the primary visual cortex, V1. In conventional sparse coding, the data are described as a combination of elementary features involving both additive and subtractive components. However, the fact that features can ‘cancel each other out’ using subtraction is contrary to the intuitive notion of combining parts to form a whole. Thus, it has recently been argued forcefully for completely non-negative representations. This paper presents Non-Negative Sparse Coding (NNSC) applied to the learning of facial features for face recognition and a comparison is made with the other part-based techniques, Non-negative Matrix Factorization (NMF) and Local-Non-negative Matrix Factorization (LNMF). The NNSC approach has been tested on the Aleix–Robert (AR), the Face Recognition Technology (FERET), the Yale B, and the Cambridge ORL databases, respectively. In doing so, we have compared and evaluated the proposed NNSC face recognition technique under varying expressions, varying illumination, occlusion with sunglasses, occlusion with scarf, and varying pose. Tests were performed with different distance metrics such as the L1-metric, L2-metric, and Normalized Cross-Correlation (NCC). All these experiments involved a large range of basis dimensions. In general, NNSC was found to be the best approach of the three part-based methods, although it must be observed that the best distance measure was not consistent for the different experiments.

Sparse Approximation of Images Inspired from the Functional Architecture of the Primary Visual Areas

Several drawbacks of critically sampled wavelets can be solved by overcomplete multiresolution transforms and sparse approximation algorithms. Facing the difficulty to optimize such nonorthogonal and nonlinear transforms, we implement a sparse approximation scheme inspired from the functional architecture of the primary visual cortex. The scheme models simple and complex cell receptive fields through log-Gabor wavelets. The model also incorporates inhibition and facilitation interactions between neighboring cells. Functionally these interactions allow to extract edges and ridges, providing an edge-based approximation of the visual information. The edge coefficients are shown sufficient for closely reconstructing the images, while contour representations by means of chains of edges reduce the information redundancy for approaching image compression. Additionally, the ability to segregate the edges from the noise is employed for image restoration.

Simple fall-off pattern of correlated neural activity in the developing lateral geniculate nucleus

Activity-dependent models for cortical simple-cell receptive field development predict specific patterns of correlated neural activity within the visual pathway, such as a Mexican hat-shaped pattern of correlated activity in the lateral geniculate nucleus (LGN). However, such activity patterns have yet to be experimentally demonstrated. We performed multielectrode recordings in the LGN of immature awake ferrets and found simple fall-off-shaped, rather than Mexican hat-shaped, patterns of correlated activity. A weak surround in the LGN neuron's receptive field and the statistics of the input contributed to this pattern of correlated activity. Computer simulation of cortical receptive field development incorporating the experimentally observed activity patterns demonstrated that a simple-cell receptive field emerges when a newly devised 'split' constraint on synaptic growth is combined with Hebbian synaptic modification rules. Thus, given certain developmental constraints on synaptic plasticity, patterns of correlated activity within the LGN are compatible with Hebbian models of simple-cell receptive field development.

ARE SPARSE-CODING SIMPLE CELL RECEPTIVE FIELD MODELS PHYSIOLOGICALLY PLAUSIBLE?

Olshausen and Field (1996) developed a simple cell receptive field model for natural scene processing in V1, based on unsupervised learning and non-orthogonal basis function optimization of an overcomplete representation of visual space. The model was originally tested with an ensemble of whitened natural scenes, simulating pre-cortical filtering in the retinal ganglia and lateral geniculate nucleus, and the basis functions qualitatively resembled the orientation-specific responses of V1 simple cells in the spatial domain. In this study, the quantitative tuning responses of the basis functions in the spectral domain are estimated using a Gaussian model, to determine their goodness-of-fit to the known bandwidths of simple cells in primate V1. Five simulation experiments which examined key features of the model are reported: changing the size of the basis functions; using a complete versus over-complete representation; changing the sparseness factor; using a variable learning rate; and mapping the basis functions with a whitening spatial function. The key finding of this study is that across all image themes, basis function sizes, number of basis functions, sparseness factors and learning rates, the spatial-frequency tuning did not closely resemble that of primate area 17 -- the model results more closely resembled the unclassified cat neurones of area 19 with a single exception, and not area 17 as predicted.

Learning receptive fields using predictive feedback

Previously, it was suggested that feedback connections from higher- to lower-level areas carry predictions of lower-level neural activities, whereas feedforward connections carry the residual error between the predictions and the actual lower-level activities [Rao, R.P.N., Ballard, D.H., 1999. Nature Neuroscience 2, 79–87.]. A computational model implementing the hypothesis learned simple cell receptive fields when exposed to natural images. Here, we use predictive feedback to explain tuning properties in medial superior temporal area (MST). We implement the hypothesis using a new, biologically plausible, algorithm based on matching pursuit, which retains all the features of the previous implementation, including its ability to efficiently encode input. When presented with natural images, the model developed receptive field properties as found in primary visual cortex. In addition, when exposed to visual motion input resulting from movements through space, the model learned receptive field properties resembling those in MST. These results corroborate the idea that predictive feedback is a general principle used by the visual system to efficiently encode natural input.

Receptive fields of simple cells from a taxonomic study of natural images and suppression of scale redundancy

Much effort has been carried out to propose models of the early visual pathway based on the statistical analysis of natural images. These conventional frameworks lead to predictions on the RFs of simple cells which do not fit well their observed properties [D.L. Ringach, Spatial structure and symmetry of simple-cell receptive fields in macaque primary visual cortex, J. Neurophysiol. 2002 (2002) 455–463]. In order to overcome these difficulties, we have been carrying a research program to derive robust coding principles from the statistics of natural images [A. Turiel, G. Mato, N. Parga, J.P. Nadal, Self-similarity properties of natural images, in: Proceedings of NIPS’97, vol. 10, MIT Press, 1997, pp. 836–842; A. Turiel, J.M. Delgado, N. Parga, Learning efficient internal representations from natural image collections, Neurocomputing 58–60 (2004) 915–921]. Two principles emerging from our study of image statistics are: first, there exists scale redundancy and this can be eliminated from the code; second, images are constructed by a combination (partly linear and partly non-linear) of some simple patterns of contrast (edges, bars, and composites of these two). These two principles can be used to derive filters which explain observed properties of the RFs of simple cells and which compare quite well with the results reported by Ringach [Spatial structure and symmetry of simple-cell receptive fields in macaque primary visual cortex, J. Neurophysiol. 2002 (2002) 455–463] and previous experimental work.

Attention Modulates the Responses of Simple Cells in Monkey Primary Visual Cortex

Spatial attention has long been postulated to act as a spotlight that increases the salience of visual stimuli at the attended location. We examined the effects of attention on the receptive fields of simple cells in primary visual cortex (V1) by training macaque monkeys to perform a task with two modes. In the attended mode, the stimuli relevant to the animal's task overlay the receptive field of the neuron being recorded. In the unattended mode, the animal was cued to attend to stimuli outside the receptive field of that neuron. The relevant stimulus, a colored pixel, was briefly presented within a white-noise stimulus, a flickering grid of black and white pixels. The receptive fields of the neurons were mapped by correlating spikes with the white-noise stimulus in both attended and unattended modes. We found that attention could cause significant modulation of the visually evoked response despite an absence of significant effects on the overall firing rates. On further examination of the relationship between the strength of the visual stimulation and the firing rate, we found that attention appears to cause multiplicative scaling of the visually evoked responses of simple cells, demonstrating that attention reaches back to the initial stages of visual cortical processing.

Spatial Structure of Complex Cell Receptive Fields Measured with Natural Images

Neuronal receptive fields (RFs) play crucial roles in visual processing. While the linear RFs of early neurons have been well studied, RFs of cortical complex cells are nonlinear and therefore difficult to characterize, especially in the context of natural stimuli. In this study, we used a nonlinear technique to compute the RFs of complex cells from their responses to natural images. We found that each RF is well described by a small number of subunits, which are oriented, localized, and bandpass. These subunits contribute to neuronal responses in a contrast-dependent, polarity-invariant manner, and they can largely predict the orientation and spatial frequency tuning of the cell. Although the RF structures measured with natural images were similar to those measured with random stimuli, natural images were more effective for driving complex cells, thus facilitating rapid identification of the subunits. The subunit RF model provides a useful basis for understanding cortical processing of natural stimuli.

DEVELOPMENT OF FEEDFORWARD RECEPTIVE FIELD STRUCTURE OF A SIMPLE CELL AND ITS CONTRIBUTION TO THE ORIENTATION SELECTIVITY: A MODELING STUDY

Recent experimental studies of hetero-synaptic interactions in various systems have shown the role of signaling in the plasticity, challenging the conventional understanding of Hebb's rule. It has also been found that activity plays a major role in plasticity, with neurotrophins acting as molecular signals translating activity into structural changes. Furthermore, role of synaptic efficacy in biasing the outcome of competition has also been revealed recently. Motivated by these experimental findings we present a model for the development of simple cell receptive field structure based on the competitive hetero-synaptic interactions for neurotrophins combined with cooperative hetero-synaptic interactions in the spatial domain. We find that with proper balance in competition and cooperation, the inputs from two populations (ON/OFF) of LGN cells segregate starting from the homogeneous state. We obtain segregated ON and OFF regions in simple cell receptive field. Our modeling study supports the experimental findings, suggesting the role of synaptic efficacy and the role of spatial signaling. We find that using this model we obtain simple cell RF, even for positively correlated activity of ON/OFF cells. We also compare different mechanism of finding the response of cortical cell and study their possible role in the sharpening of orientation selectivity. We find that degree of selectivity improvement in individual cells varies from case to case depending upon the structure of RF field and type of sharpening mechanism.