Gaussian Receptive Fields Research Articles

Design of a Neuronal Array

Retinal ganglion cells of a given type overlap their dendritic fields such that every point in space is covered by three to four cells. We investigated what function is served by such extensive overlap. Recording from pairs of ON or OFF brisk-transient ganglion cells at photopic intensities, we confirmed that this overlap causes the Gaussian receptive field centers to be spaced at approximately 2 SDs (sigma). This, together with response nonlinearities and variability, was just sufficient to provide an ideal observer with uniform contrast sensitivity across the retina for both threshold and suprathreshold stimuli. We hypothesized that overlap might maximize the information represented from natural images, thereby optimizing retinal performance for many tasks. Indeed, tested with natural images (which contain statistical correlations), a model ganglion cell array maximized information represented in its population responses with approximately 2sigma spacing, i.e., the overlap observed in the retina. Yet, tested with white noise (which lacks statistical correlations), an array maximized its information by minimizing overlap. In both cases, optimal overlap balanced greater signal-to-noise ratio (from larger receptive fields) against greater redundancy (because of larger receptive field overlap). Thus, dendritic overlap improves vision by taking optimal advantage of the statistical correlations of natural scenes.

Aprendizado não-supervisionado em redes neurais pulsadas de base radial: um estudo da capacidade de agrupamento para a classificação de pixels

Redes neurais pulsadas - redes que utilizam uma codificação temporal da informação - têm despontado como uma promissora abordagem dentro do paradigma conexionista, emergente da ciência cognitiva. Um desses novos modelos é a rede neural pulsada com função de base radial, que é capaz de armazenar informação nos tempos de atraso axonais dos neurônios. Um algoritmo de aprendizado foi aplicado com sucesso nesta rede pulsada, que se mostrou capaz de mapear uma seqüência de pulsos de entrada em uma seqüência de pulsos de saída. Mais recentemente, um método baseado no uso de campos receptivos gaussianos foi proposto para codificar dados constantes em uma seqüência de pulsos temporais. Este método tornou possível a essa rede lidar com dados computacionais. O processo de aprendizado desta nova rede não se encontra plenamente compreendido e investigações mais profundas são necessárias para situar este modelo dentro do contexto do aprendizado de máquinas e também para estabelecer as habilidades e limitações desta rede. Este trabalho apresenta uma investigação desse novo classificador e um estudo de sua capacidade de agrupar dados em três dimensões, particularmente procurando estabelecer seus domínios de aplicação e horizontes no campo da visão computacional.

Simultaneous constraints on pre- and post-synaptic cells couple cortical feature maps in a 2D geometric model of orientation preference

The most prominent feature of mammalian striate cortex (V1) is the spatial organization of response preferences for the position and orientation of elementary visual stimuli. Models for the formation of cortical maps of orientation and 'retinotopic' position typically rely on a combination of Hebbian or correlation-based synaptic plasticity, and constraints on the distribution of synaptic weights. We consider a simplified model of orientation and retinotopic specificity based on the geometry of the feed-forward synaptic weight distribution from an 'unoriented' layer of cells to a first weakly oriented layer. We model the feed-forward weight distribution as a system of planar Gaussian receptive fields each elongated in the direction matching the preferred orientation of the postsynaptic cell. Under the constraint of presynaptic weight normalization (each cell in the oriented layer receives the same net synaptic weight) and a uniform retinotopic map (displacement of centres of mass of receptive fields in the unoriented layer is strictly proportional to the displacement of the corresponding cells in the oriented layer), we find that imposing a pattern of orientation preference forces the system to violate postsynaptic weight normalization (each cell in the unoriented layer no longer sends forth the same net synaptic weight). We study this deviation from uniformity of the postsynaptic weight, and find that the deviation has a distinct form in the vicinity of the 'pinwheel' singularities of the orientation map. We show that uniform synaptic coverage of the unoriented layer can be restored by introducing a distortion in the retinotopic locations of the receptive fields. We calculate, to first order in the relative elongation of the receptive fields, the retinotopic distortion vector field. Both the pattern of postsynaptic weight non-uniformity and the corrective retinotopic distortion vector field fail to possess the reflection symmetry commonly assumed to relate orientation singularities with topological index +/- pi. Hence, we show that 'right-handed' and 'left-handed' orientation singularities are funda-mentally distinct anatomical structures when full 2D synaptic architecture is taken into account. Finally, we predict specific patterns of retinotopic distortion that should obtain in the vicinity of +/- pi-fold orientation singularities, if uniform pre- and post-synaptic weight constraints are strongly enforced.

Brand identification using Gaussian derivative histograms

In this article, we describe a module for the identification of brand logos from video data. A model for the visual appearance of each logo is generated from a small number of sample images using multidimensional histograms of scale-normalized chromatic Gaussian receptive fields. We compare several identification techniques based on multidimensional histograms. Each of the methods displays high recognition rates and can be used for logo identification. Our method for calculating scale-normalized Gaussian receptive fields has linear computational complexity and is thus well adapted to a real-time system. However, with the current generation of microprocessors we obtain at best only two images per second when processing a full PAL video stream. To accelerate the process, we propose an architecture that combines fast detection, reliable identification, and fast tracking for speedup. The resulting real-time system is evaluated using video streams from sports Formula 1 races and football.

Temporal visual filtering in diabetes mellitus

The background modulation method was used to investigate the temporal response of the magnocellular pathway in diabetic patients and controls. The luminance threshold for detecting a moving, 2°, achromatic target was measured as a function of background flicker frequency from 5 to 45 Hz. A model of photoreceptor kinetics integrated with difference of Gaussian receptive fields [Vis. Neurosci. 13 (1996) 173] was used to analyse the data. Diabetic patients with significant maculopathy showed raised thresholds at 8.75, 12.5, 15 and 17.5 Hz. Estimates of photoreceptor summation time were the same in both groups, but receptive field centre-to-surround delay showed an increasing trend in the diabetic patients.

A parallel analog intelligent vision sensor with a variable receptive field

The retina is an intelligent vision sensor indispensable in real-time image information processing of animals. The parallel image processing function of the retina was realized in an analog CMOS integrated circuit (vision chip). The chip was implemented with a one-dimensional Laplacian–Gaussian receptive field the size of which can be controlled by external signals. Active pixel sensors were used in the sensor part of the chip where the light sensitivity can be controlled by varying the accumulation time. Circuits compensating for noise due to the mismatch of transistor characteristics were incorporated, and the experimental results verified that sufficiently accurate outputs can be obtained under natural illumination. The chip can effectively extract the spatial frequency bandwidth of the signal in the presence of noise. © 1999 Scripta Technica, Syst Comp Jpn, 30(1): 60–69, 1999

A parallel analog intelligent vision sensor with a variable receptive field

The retina is an intelligent vision sensor indispensable in real-time image information processing of animals. The parallel image processing function of the retina was realized in an analog CMOS integrated circuit (vision chip). The chip was implemented with a one-dimensional Laplacian–Gaussian receptive field the size of which can be controlled by external signals. Active pixel sensors were used in the sensor part of the chip where the light sensitivity can be controlled by varying the accumulation time. Circuits compensating for noise due to the mismatch of transistor characteristics were incorporated, and the experimental results verified that sufficiently accurate outputs can be obtained under natural illumination. The chip can effectively extract the spatial frequency bandwidth of the signal in the presence of noise. © 1999 Scripta Technica, Syst Comp Jpn, 30(1): 60–69, 1999

Topographical mapping forms of objects into gaussian elastic net

A gaussian elastic net for modeling visual processing of object forms is proposed. The mechanism of the model is based on a cognitive fact that a form of a geometrical object is characterised by borders of the object. Gaussian receptive fields are used as templates which are activated by local detection of borders. The gaussian receptive fields are topological arranged such that a gross global shape of the object is mapped into the retina. By coupling gaussian receptive fields to the nodes of an elastic net, which adaptively matches the form of an object to the nodes, the gross global shape of an object is sharpened by the net as cortical processing. To achieve this, an energy function is defined. The energy function depends on the neighbourhood distances of the nodes and the activation of the gaussian receptive fields coupling to the nodes of the net. The object form is mapped into the net and sharpened by adaptively modifying the energy function to reach an optimal value. Simulation was done with an irregular object to show the processing of the net for an arbitrary object.

Neural networks for maximum likelihood clustering

Abstract Winner-take-all algorithms are commonly used techniques in clustering analysis. However, they have some problems ranging from clusters under utilization to the extended training time. Some solutions to these problems are addressed here. It is shown here that using the maximum-likelihood criterion instead of the Euclidean distance metric results in better clustering. The clusters are represented by a set of neuron each has a Gaussian receptive field. For these Gaussian neurons, the covariance matrices, in addition to the centers, are learned. The one-winner condition is relaxed by maximizing the likelihood function of the mixture density function of the samples. This produces larger likelihood values and more normally distributed clusters. A fast mixture likelihood clustering is provided for both batch and pattern learning modes. Convergence analysis and experimental results are also presented.

Can the data of Campbell and Robson be explained without assuming Fourier analysis?

Certain experiments on the detection of low-contrast gratings, occasionally cited as evidence of Fourier analysis within the visual system, are interpreted without the assumption of Fourier analysis. Theoretical curves are obtained and compared with the published experimental points, showing mostly satisfactory agreement. The computations utilize Gaussian receptive fields (on-center and off-center) for the retinal ganglion cells, spatial summation, center-surround antagonism, quasilinear response at low contrasts (X-cells), and the assumption that the first significant convergence is primarily between cells of like response type and like receptive field geometry.