The Gaussian derivative model for spatial-temporal vision: II. Cortical data

Receptive fields of simple cells in the primate visual cortex were well fit in the space and time domains by the Gaussian Derivative (GD) model for spatio-temporal vision. All 23 fields in the data sample could be fit by one equation. varying only a single shape number and nine geometric transformation parameters. A difference-of-offset-Gaussians (DOOG) mechanism for the GD model also fit the data well. Other models tested did not fit the data as well as or as succinctly, or failed to converge on a unique solution, indicating over-parameterization. An efficient computational algorithm was found for the GD model which produced robust estimates of the direction and speed of moving objects in real scenes.     

The Gaussian derivative model for spatial-temporal vision: I. Cortical model

How do we see the motion of objects as well as their shapes? The Gaussian Derivative (GD) spatial model is extended to time to help answer this question. The GD spatio-temporal model requires only two numbers to describe the complete three-dimensional space-time shapes of individual receptive fields in primate visual cortex. These two numbers are the derivative numbers along the respective spatial and temporal principal axes of a given receptive field. Nine transformation parameters allow for a standard geometric association of these intrinsic axes with the extrinsic environment. The GD spatio-temporal model describes in one framework the following properties of primate simple cell fields: motion properties, number of lobes in space-time, spatial orientation. location, and size. A discrete difference-of-offset-Gaussians (DOOG) model provides a plausible physiological mechanism to form GD-like model fields in both space and time. The GD model hypothesizes that receptive fields at the first stage of processing in the visual cortex approximate 'derivative analyzers' that estimate local spatial and temporal derivatives of the intensity profile in the visual environment. The receptive fields as modeled provide operators that can allow later stages of processing in either a biological or machine vision system to estimate the motion as well as the shapes of objects in the environment.     

Factors limiting peripheral pattern discrimination

Previous reports indicate that some foveally discriminable compound gratings are indiscriminable in peripheral vision, even when they are scaled by the ratio of peripheral to foveal grating acuity. To determine the stimulus properties that limit peripheral discrimination, we used Gaussian derivatives of various orders. These patterns are spatially localized and have intrinsic even or odd symmetry. Our results show that certain odd symmetric patterns are discriminable in the periphery, while others are not. Furthermore, certain even symmetric patterns are not peripherally discriminable. These data are consistent with three limitations on peripheral pattern discrimination: (1) Patterns that produce different maximum neural responses will be peripherally discriminable. (2) Positional uncertainty and undersampling degrade discrimination of high spatial frequency patterns in the periphery. (3) Patterns generating substantial neural activity within a constrained region are processed as textures in peripheral vision so that pattern details within that region are no longer available for discrimination. A neural model incorporating inhibition of simple cells by complex cells implements a transition between contour analysis and texture analysis in peripheral vision and explains the experimental data.     

Patterns of chromatic information processing in the lobula of the honeybee, Apis mellifera L

The honeybee, Apis mellifera L., is one of the living creatures that has its colour vision proven through behavioural tests. Previous studies of honeybee colour vision has emphasized the relationship between the spectral sensitivities of photoreceptors and colour discrimination behaviour. The current understanding of the neural mechanisms of bee colour vision is, however, rather limited. The present study surveyed the patterns of chromatic information processing of visual neurons in the lobula of the honeybee, using intracellular recording stimulated by three light-emitting diodes, whose emission spectra approximately match the spectral sensitivity peaks of the honeybee. The recorded visual neurons can be divided into two groups: non-colour opponent cells and colour opponent cells. The non-colour opponent cells comprise six types of broad-band neurons and four response types of narrow-band neurons. The former might detect brightness of the environment or function as chromatic input channels, and the latter might supply specific chromatic input. Amongst the colour opponent cells, the principal neural mechanism of colour vision, eight response types were recorded. The receptive fields of these neurons were not centre surround as observed in primates. Some recorded neurons with tonic post-stimulus responses were observed, however, suggesting temporal defined spectral opponency may be part of the colour-coding mechanisms.     

线画图形质地辨认的一个神经网络模型

本文提出了一个能辨认线画质地的神经网络模型,并在IBM-PC机上对它进行了成功的模拟.该模型可与真实的视觉系统建立起很好的对应关系,因此可以认为它是对视觉系统质地辨认过程的一个机器再现.另外,本文推广了原广义Gabor函数模型,得到了一个适合于描写网络层次视觉功能的单细胞感受野的数学描述.     

Effect of sizes of the retinal receptive field on the precision of image perception

A paradox of sharp vision regardless of relatively large receptive fields of the retina is discussed. Transformation of images by the receptive fields or intersecting fields of vision of omma-tidia of the complex eye is described as an approximation of Bool functions by the implicant forms. Such an approximation results in the information compression. The compression coefficient is determined by the ratio between the implicants number and the number of points in the field of determination of the Bool function. The approximation errors strongly depend on the size of the receptive fields. With their increase the error abruptly decreases and only after passing the deep minimum it begins to slowly increase. The mechanism of information compression is a universal one and is similarly realized on the retina, facet and t.v. sets.     

The role of retinal bipolar cell in early vision: an implication with analogue networks and regularization theory

A linear analogue network model is proposed to describe the neuronal circuit of the outer retina consisting of cones, horizontal cells, and bipolar cells. The model reflects previous physiological findings on the spatial response properties of these neurons to dim illumination and is expressed by physiological mechanisms, i.e., membrane conductances, gap-junctional conductances, and strengths of chemical synaptic interactions. Using the model, we characterized the spatial filtering properties of the bipolar cell receptive field with the standard regularization theory, in which the early vision problems are attributed to minimization of a cost function. The cost function accompanying the present characterization is derived from the linear analogue network model, and one can gain intuitive insights on how physiological mechanisms contribute to the spatial filtering properties of the bipolar cell receptive field. We also elucidated a quantitative relation between the Laplacian of Gaussian operator and the bipolar cell receptive field. From the computational point of view, the dopaminergic modulation of the gap-junctional conductance between horizontal cells is inferred to be a suitable neural adaptation mechanism for transition between photopic and mesopic vision. Received: 20 December 1996 / Accepted in revised form: 16 May 1997     

Decoding successive computational stages of saliency processing

An important requirement for vision is to identify interesting and relevant regions of the environment for further processing. Some models assume that salient locations from a visual scene are encoded in a dedicated spatial saliency map [1, 2]. Then, a winner-take-all (WTA) mechanism [1, 2] is often believed to threshold the graded saliency representation and identify the most salient position in the visual field. Here we aimed to assess whether neural representations of graded saliency and the subsequent WTA mechanism can be dissociated. We presented images of natural scenes while subjects were in a scanner performing a demanding fixation task, and thus their attention was directed away. Signals in early visual cortex and posterior intraparietal sulcus (IPS) correlated with graded saliency as defined by a computational saliency model. Multivariate pattern classification [3, 4] revealed that the most salient position in the visual field was encoded in anterior IPS and frontal eye fields (FEF), thus reflecting a potential WTA stage. Our results thus confirm that graded saliency and WTA-thresholded saliency are encoded in distinct neural structures. This could provide the neural representation required for rapid and automatic orientation toward salient events in natural environments.     

A statistical analysis of natural images matches psychophysically derived orientation tuning curves.

A neural net method is used to extract principal components from real-world images. The initial components are a Gaussian followed by horizontal and vertical operators, starting with the first derivative and moving to successively higher orders. Two of the components are 'bar-detectors'. Their measured orientation selectivity is similar to that suggested by Foster & Ward (Proc. R. Soc. Lond. B 243, 75 (1991] to account for brief-exposure psychophysical data. In tests with noise images, the ratio of sensitivity between the two components is controlled by the degree of anisotropy in the image.     

Investigation of the mechanisms for wireless nerve stimulation without active electrodes

Electric-field stimulation of neuronal activity can be used to improve the speed of regeneration for severed and damaged nerves. Most techniques, however, require invasive electronic circuitry which can be uncomfortable for the patient and can damage surrounding tissue. A recently suggested technique uses a graft-antenna—a metal ring wrapped around the damaged nerve—powered by an external magnetic stimulation device. This technique requires no electrodes and internal circuitry with leads across the skin boundary or internal power, since all power is provided wirelessly. This paper examines the microscopic basic mechanisms that allow the magnetic stimulation device to cause neural activation via the graft-antenna. A computational model of the system was created and used to find that under magnetic stimulation, diverging electric fields appear at the metal ring's edges. If the magnetic stimulation is sufficient, the gradients of these fields can trigger neural activation in the nerve. In-vivo measurements were also performed on rat sciatic nerves to support the modeling finding that direct contact between the antenna and the nerve ensures neural activation given sufficient magnetic stimulation. Simulations also showed that the presence of a thin gap between the graft-antenna and the nerve does not preclude neural activation but does reduce its efficacy.     

The membrane potential process of a single neuron seen as a cumulative damage process

A simple integrate-and-fire mechanism of a single neuron can be compared with a cumulative damage process, where the spiking process is analogous to rupture sequences of a material under cycles of stress. Although in some cases lognormal-like patterns can be recognized in the inter-spike times under a simple integrate-and-fire mechanism, fatigue life models as the inverse Gaussian distribution and the Birnbaum–Saunders distribution (which was recently introduced in the neural activity framework) provide theoretical arguments that make them more suitable for the modeling of the resulting inter-spike times.     

Constructive Autoassociative Neural Network for Facial Recognition

Autoassociative artificial neural networks have been used in many different computer vision applications. However, it is difficult to define the most suitable neural network architecture because this definition is based on previous knowledge and depends on the problem domain. To address this problem, we propose a constructive autoassociative neural network called CANet (Constructive Autoassociative Neural Network). CANet integrates the concepts of receptive fields and autoassociative memory in a dynamic architecture that changes the configuration of the receptive fields by adding new neurons in the hidden layer, while a pruning algorithm removes neurons from the output layer. Neurons in the CANet output layer present lateral inhibitory connections that improve the recognition rate. Experiments in face recognition and facial expression recognition show that the CANet outperforms other methods presented in the literature.     

Construction of complex receptive fields in cat primary visual cortex.

In primary visual cortex, neurons are classified into simple cells and complex cells based on their response properties. Although the role of these two cell types in vision is still unknown, an attractive hypothesis is that simple cells are necessary to construct complex receptive fields. This hierarchical model puts forward two main predictions. First, simple cells should connect monosynaptically to complex cells. Second, complex cells should become silent when simple cells are inactivated. We have recently provided evidence for the first prediction, and here we do the same for the second. In summary, our results suggest that the receptive fields of most layer 2+3 complex cells are generated by a mechanism that requires simple cell inputs.     

A computational theory of visual receptive fields

A receptive field constitutes a region in the visual field where a visual cell or a visual operator responds to visual stimuli. This paper presents a theory for what types of receptive field profiles can be regarded as natural for an idealized vision system, given a set of structural requirements on the first stages of visual processing that reflect symmetry properties of the surrounding world. These symmetry properties include (i) covariance properties under scale changes, affine image deformations, and Galilean transformations of space–time as occur for real-world image data as well as specific requirements of (ii) temporal causality implying that the future cannot be accessed and (iii) a time-recursive updating mechanism of a limited temporal buffer of the past as is necessary for a genuine real-time system. Fundamental structural requirements are also imposed to ensure (iv) mutual consistency and a proper handling of internal representations at different spatial and temporal scales. It is shown how a set of families of idealized receptive field profiles can be derived by necessity regarding spatial, spatio-chromatic, and spatio-temporal receptive fields in terms of Gaussian kernels, Gaussian derivatives, or closely related operators. Such image filters have been successfully used as a basis for expressing a large number of visual operations in computer vision, regarding feature detection, feature classification, motion estimation, object recognition, spatio-temporal recognition, and shape estimation. Hence, the associated so-called scale-space theory constitutes a both theoretically well-founded and general framework for expressing visual operations. There are very close similarities between receptive field profiles predicted from this scale-space theory and receptive field profiles found by cell recordings in biological vision. Among the family of receptive field profiles derived by necessity from the assumptions, idealized models with very good qualitative agreement are obtained for (i) spatial on-center/off-surround and off-center/on-surround receptive fields in the fovea and the LGN, (ii) simple cells with spatial directional preference in V1, (iii) spatio-chromatic double-opponent neurons in V1, (iv) space–time separable spatio-temporal receptive fields in the LGN and V1, and (v) non-separable space–time tilted receptive fields in V1, all within the same unified theory. In addition, the paper presents a more general framework for relating and interpreting these receptive fields conceptually and possibly predicting new receptive field profiles as well as for pre-wiring covariance under scaling, affine, and Galilean transformations into the representations of visual stimuli. This paper describes the basic structure of the necessity results concerning receptive field profiles regarding the mathematical foundation of the theory and outlines how the proposed theory could be used in further studies and modelling of biological vision. It is also shown how receptive field responses can be interpreted physically, as the superposition of relative variations of surface structure and illumination variations, given a logarithmic brightness scale, and how receptive field measurements will be invariant under multiplicative illumination variations and exposure control mechanisms.     

Characteristics of the background fields in multichannel-recorded magnetic field responses

We have studied the background fields in the auditory evoked magnetic field responses recorded with a 37-channel SQUID magnetometer. The background fields were found to have a main contribution from the spontaneous fields, which originate in the neural activities of the brain. The spontaneous fields had strong spatial correlation across the recording sites even after averaging over 100 epochs. The spatial distribution of the spontaneous field consisted of 3 main components of single extremum pattern, dipolar pattern, and a dipolar pattern with some distortion. Computer simulations of the localization of a single dipole source of the evoked field response showed that the spontaneous background field could bring about large location errors in an unpredictable manner, as compared with the location errors caused by a spatially random Gaussian noise field.     

An adaptive visual neuronal model implementing competitive, temporally asymmetric Hebbian learning

A novel depth-from-motion vision model based on leaky integrate-and-fire (I&F) neurons incorporates the implications of recent neurophysiological findings into an algorithm for object discovery and depth analysis. Pulse-coupled I&F neurons capture the edges in an optical flow field and the associated time of travel of those edges is encoded as the neuron parameters, mainly the time constant of the membrane potential and synaptic weight. Correlations between spikes and their timing thus code depth in the visual field. Neurons have multiple output synapses connecting to neighbouring neurons with an initial Gaussian weight distribution. A temporally asymmetric learning rule is used to adapt the synaptic weights online, during which competitive behaviour emerges between the different input synapses of a neuron. It is shown that the competition mechanism can further improve the model performance. After training, the weights of synapses sourced from a neuron do not display a Gaussian distribution, having adapted to encode features of the scenes to which they have been exposed.     

Visual Perception-Based Statistical Modeling of Complex Grain Image for Product Quality Monitoring and Supervision on Assembly Production Line

Computer vision as a fast, low-cost, noncontact, and online monitoring technology has been an important tool to inspect product quality, particularly on a large-scale assembly production line. However, the current industrial vision system is far from satisfactory in the intelligent perception of complex grain images, comprising a large number of local homogeneous fragmentations or patches without distinct foreground and background. We attempt to solve this problem based on the statistical modeling of spatial structures of grain images. We present a physical explanation in advance to indicate that the spatial structures of the complex grain images are subject to a representative Weibull distribution according to the theory of sequential fragmentation, which is well known in the continued comminution of ore grinding. To delineate the spatial structure of the grain image, we present a method of multiscale and omnidirectional Gaussian derivative filtering. Then, a product quality classifier based on sparse multikernel–least squares support vector machine is proposed to solve the low-confidence classification problem of imbalanced data distribution. The proposed method is applied on the assembly line of a food-processing enterprise to classify (or identify) automatically the production quality of rice. The experiments on the real application case, compared with the commonly used methods, illustrate the validity of our method.     

Color perception is mediated by a plastic neural mechanism that is adjustable in adults

An intensely debated issue concerning visual-experience-dependent neural plasticity is whether experience is required only to maintain function or whether information from experience is used actively, relieving the necessity to hard-wire all connections and allowing adaptive adjustments. Here, an active role for experience is demonstrated in circuits for color vision. Chromatic experience was altered using colored filters. Over days there was a shift in color perception, as measured by the wavelength of unique yellow, which persisted 1-2 weeks after the filters were discontinued. Moreover, color-deficient adults were shown to have altered weightings of inputs to chromatic channels, demonstrating a large neural adjustment to their inherited photopigment defect. Thus, a neural normalization mechanism for color perception, determined by visual experience, operates to compensate for large genetic differences in retinal architecture and for changes in chromatic environment.     

Asymmetry in neural fields: a spatiotemporal encoding mechanism

Neural field models have been successfully applied to model diverse brain mechanisms like visual attention, motor control, and memory. Most theoretical and modeling works have focused on the study of the dynamics of such systems under variations in neural connectivity, mainly symmetric connectivity among neurons. However, less attention has been given to the emerging properties of neuron populations when neural connectivity is asymmetric, although asymmetric activity propagation has been observed in cortical tissue. Here we explore the dynamics of neural fields with asymmetric connectivity and show, in the case of front propagation, that it can bias the population to follow a certain trajectory with higher activation. We find that asymmetry relates linearly to the input speed when the input is spatially localized, and this relation holds for different kernels and input shapes. To illustrate the behavior of asymmetric connectivity, we present an application: standard video sequences of human motion were encoded using the asymmetric neural field and compared to computer vision techniques. Overall, our results indicate that asymmetric neural fields are a competitive approach for spatiotemporal encoding with two main advantages: online classification and distributed operation.