A general model for visual motion detection

We propose a general model for detection of both first-order motion and second-order motion. In this model an input stimulus is divided into a number of partially overlapping spatiotemporal local regions. Spatiotemporal frequency analysis is done for every local region using Gabor filters, then the input stimulus (original spatiotemporal signal) is replaced by the outputs of Gabor filters. Local motion is detected by applying Gabor motion detectors to each local spatiotemporal pattern depicted by each local feature value. Outputs of all the detectors are integrated to give the final output for global motion of the input stimulus. The model was simulated on a computer and was confirmed to correctly detect second-order motion as well as first-order motion.     

A neural network model for stroboscopic alternative motion

A neural network which models multistable perception is presented. The network consists of sensor and inner neurons. The dynamics is established by a stochastic neuronal dynamics, a formal Hebb-type coupling dynamics and a resource mechanism that corresponds to saturation effects in perception. From this a system of coupled differential equations is derived and analyzed. Single stimuli are bound to exactly one percept, even in ambiguous situations where multistability occurs. The network exhibits discontinuous as well as continuous phase transitions and models various empirical findings, including the percepts of succession, alternative motion and simultaneity; the percept of oscillation is explained by oscillating percepts at a continuous phase transition. Received: 13 September 1995 / Accepted: 3 June 1996     

A biologically plausible acoustic motion detection neural network

In this paper we present an acoustic motion detection system to be used in a small mobile robot. While the first purpose of the system has been to be a reliable computational implementation, cheap enough to be built in hardware, effort has also been taken to construct a biologically plausible solution. The motion detector consists of a neural network composed of motion-direction sensitive neurons with a preferred direction and a preferred region of the azimuth. The system was designed to produce a higher response when stimulated by motion in the preferred direction than in the null direction and that is in fact what the system does, which means that, as desired, the system can detect motion and distinguish its direction.     

A multilayer neural network model for perception of rotational motion

A multilayer neural nerwork model for the perception of rotational motion has been developed usingReichardt's motion detector array of correlation type, Kohonen's self-organized feature map and Schuster-Wagner's oscillating neural network. It is shown that the unsupervised learning could make the neurons on the second layer of the network tend to be self-organized in a form resembling columnar organization of selective directions in area MT of the primate's visual cortex. The output layer can interpret rotation information and give the directions and velocities of rotational motion. The computer simulation results are in agreement with some psychophysical observations of rotation-al perception. It is demonstrated that the temporal correlation between the oscillating neurons would be powerful for solving the "binding problem" of shear components of rotational motion.     

A neural network model for selective attention in visual pattern recognition

A neural network model of the mechanism of selective attention in visual pattern recognition is proposed and simulated on a digital computer.When a complex figure consisting of two patterns or more is presented to the model, it is segmented into individual patterns, and each pattern is recognized separately. Even if one of the patterns to which the model is paying selective attention is affected by noise or defects, the model can recall the complete pattern from which the noise has been eliminated and the defects corrected. It is not necessary for perfect recall that the stimulus pattern should be identical in shape to the training pattern. Even though the pattern is distorted in shape or changed in size, it can be correctly recognized and the missing portions restored.The model consists of a hierarchical neural network which has efferent as well as afferent connections between cells. The afferent and the efferent signals interact with each other in the network: the efferent signals, that is, the signals for selective attention, have a facilitating effect on the afferent ones, and, at the same time, the afferent signals gate efferent signal flow. When some feature in the stimulus is not extracted in the afferent paths, the threshold for detection of that feature is automatically lowered by decreasing the efficiency of inhibition, and the model tries to extract even vague traces of the undetected feature.     

A neural net model for masking phenomena

Some aspects of masking phenomena are considered in terms of the simplest possible model of two-factor neural elements. The effect of a number of variables can be accounted for, but the introduction of an internuncial element results in a masking function which need not be symmetric about zero delay interval. As an illustration, the results for a special case are compared with available data. In general, such a model results in a masking function which depends on the intensity, area, and duration of the stimuli, as well as on the temporal and spatial separation between them.     

Underlying neural computations for some visual phenomena

In this paper we examine how a large array of neurons, and their associated neural circuitry, may determine known receptive field profile types and some well-known visual phenomena including Mach bands, edge enhancement, and visual masking of one signal by another. The neural model has a spatio-temporal structure and is described by a nonlinear integropartial differential difference equation with an isotropic Gabor kernel — a Gaussian apertured cosine modulation. Several simulations are presented.This project was partially funded by Grants A4345 to M. N. Ouztöreli and A2568 to T. M. Caelli from the Natural Sciences and Engineering Research Council of Canada through the University of Alberta     

A neural network model for the development of direction selectivity in the visual cortex

A neural network model is proposed to explain the development of direction selectivity of cortical cells. The model is constructed under the following three hypotheses that are very plausible from recent neurophysiological findings. (1) Direction selectivity is developed by modifiable inhibitory synapses. (2) It results not from the direct convergence of many excitatory inputs from LGN cells but from cortical neural networks. (3) Direction-selective mechanism is independent of orientation-selective mechanism.—The model was simulated on a computer for a few kinds of inhibitory connections and initial conditions. The results were consistent with neurophysiological facts not only for normal cats but for cats reared in an abnormal visual environment.     

A neural network model for trace conditioning

We studied the dynamics of a neural network that has both recurrent excitatory and random inhibitory connections. Neurons started to become active when a relatively weak transient excitatory signal was presented and the activity was sustained due to the recurrent excitatory connections. The sustained activity stopped when a strong transient signal was presented or when neurons were disinhibited. The random inhibitory connections modulated the activity patterns of neurons so that the patterns evolved without recurrence with time. Hence, a time passage between the onsets of the two transient signals was represented by the sequence of activity patterns. We then applied this model to represent the trace eye blink conditioning, which is mediated by the hippocampus. We assumed this model as CA3 of the hippocampus and considered an output neuron corresponding to a neuron in CA1. The activity pattern of the output neuron was similar to that of CA1 neurons during trace eye blink conditioning, which was experimentally observed.     

A neural network model for texture discrimination

A model of texture discrimination in visual cortex was built using a feedforward network with lateral interactions among relatively realistic spiking neural elements. The elements have various membrane currents, equilibrium potentials and time constants, with action potentials and synapses. The model is derived from the modified programs of MacGregor (1987). Gabor-like filters are applied to overlapping regions in the original image; the neural network with lateral excitatory and inhibitory interactions then compares and adjusts the Gabor amplitudes in order to produce the actual texture discrimination. Finally, a combination layer selects and groups various representations in the output of the network to form the final transformed image material. We show that both texture segmentation and detection of texture boundaries can be represented in the firing activity of such a network for a wide variety of synthetic to natural images. Performance details depend most strongly on the global balance of strengths of the excitatory and inhibitory lateral interconnections. The spatial distribution of lateral connective strengths has relatively little effect. Detailed temporal firing activities of single elements in the lateral connected network were examined under various stimulus conditions. Results show (as in area 17 of cortex) that a single element's response to image features local to its receptive field can be altered by changes in the global context.     

A neural network model for cognitive activity

A consideration of the storage of information as an energized neuronal state leads to the development of a new type of neural network model which is capable of pattern recognition, concept formation and recognition of patterns of events in time. The network consists of several layers of cells, each cell representing by connections from the lower levels some combination of features or concepts. Information travels toward higher layers by such connections during an association phase, and then reverses during a recognition phase, where higher-order concepts can redirect the flow to more appropriate elements, revising the perception of the environment. This permits a more efficient method of distinguishing closely-related patterns and also permits the formation of negative associations, which is a likely requirement for formation of "abstract" concepts.     

An associative memory that can form hypotheses: a phase-coded neural network

 Nonlinear associative memories as realized, e.g., by Hopfield nets are characterized by attractor-type dynamics. When fed with a starting pattern, they converge to exactly one of the stored patterns which is supposed to be most similar. These systems cannot render hypotheses of classification, i.e., render several possible answers to a given classification problem. Inspired by von der Malsburg’s correlation theory of brain function, we extend conventional neural network architectures by introducing additional dynamical variables. Assuming an oscillatory time structure of neural firing, i.e., the existence of neural clocks, we assign a so-called phase to each formal neuron. The phases explicitly describe detailed correlations of neural activities neglected in conventional neural network architectures. Implementing this extension into a simple self-organizing network based on a feature map, we present an associative memory that actually is capable of forming hypotheses of classification. Received: 6 December 1993/Accepted in revised form: 14 July 1994     

A hierarchical neural network model for associative memory

A hierarchical neural network model with feedback interconnections, which has the function of associative memory and the ability to recognize patterns, is proposed. The model consists of a hierarchical multi-layered network to which efferent connections are added, so as to make positive feedback loops in pairs with afferent connections. The cell-layer at the initial stage of the network is the input layer which receives the stimulus input and at the same time works as an output layer for associative recall. The deepest layer is the output layer for pattern-recognition. Pattern-recognition is performed hierarchically by integrating information by converging afferent paths in the network. For the purpose of associative recall, the integrated information is again distributed to lower-order cells by diverging efferent paths. These two operations progress simultaneously in the network. If a fragment of a training pattern is presented to the network which has completed its self-organization, the entire pattern will gradually be recalled in the initial layer. If a stimulus consisting of a number of training patterns superposed is presented, one pattern gradually becomes predominant in the recalled output after competition between the patterns, and the others disappear. At about the same time when the recalled pattern reaches a steady state in he initial layer, in the deepest layer of the network, a response is elicited from the cell corresponding to the category of the finally-recalled pattern. Once a steady state has been reached, the response of the network is automatically extinguished by inhibitory signals from a steadiness-detecting cell. If the same stimulus is still presented after inhibition, a response for another pattern, formerly suppressed, will now appear, because the cells of the network have adaptation characteristics which makes the same response unlikely to recur. Since inhibition occurs repeatedly, the superposed input patterns are recalled one by one in turn.     

A neural network model for limb trajectory formation

This paper deals with the problem of representing and generating unconstrained aiming movements of a limb by means of a neural network architecture. The network produced time trajectories of a limb from a starting posture toward targets specified by sensory stimuli. Thus the network performed a sensory-motor transformation. The experimenters trained the network using a bell-shaped velocity profile on the trajectories. This type of profile is characteristic of most movements performed by biological systems. We investigated the generalization capabilities of the network as well as its internal organization. Experiments performed during learning and on the trained network showed that: (i) the task could be learned by a three-layer sequential network; (ii) the network successfully generalized in trajectory space and adjusted the velocity profiles properly; (iii) the same task could not be learned by a linear network; (iv) after learning, the internal connections became organized into inhibitory and excitatory zones and encoded the main features of the training set; (v) the model was robust to noise on the input signals; (vi) the network exhibited attractor-dynamics properties; (vii) the network was able to solve the motorequivalence problem. A key feature of this work is the fact that the neural network was coupled to a mechanical model of a limb in which muscles are represented as springs. With this representation the model solved the problem of motor redundancy.     

Formalism for the neural network of visual systems

A formalism to describe neural interrelations is developed on the exemplary case of the fly visual system. Absolute and relative indices are employed to identify the position of neural elements within the lattices of the visual ganglia. Illustrative applications as the projection of fly retinual cell axons into the lamina are discussed as well as the general feasibility of the formalism to other visual systems.     

A space-time delay neural network for motion recognition and its application to lipreading

Motion recognition has received increasing attention in recent years owing to heightened demand for computer vision in many domains, including the surveillance system, multimodal human computer interface, and traffic control system. Most conventional approaches classify the motion recognition task into partial feature extraction and time-domain recognition subtasks. However, the information of motion resides in the space-time domain instead of the time domain or space domain independently, implying that fusing the feature extraction and classification in the space and time domains into a single framework is preferred. Based on this notion, this work presents a novel Space-Time Delay Neural Network (STDNN) capable of handling the space-time dynamic information for motion recognition. The STDNN is unified structure, in which the low-level spatiotemporal feature extraction and high-level space-time-domain recognition are fused. The proposed network possesses the spatiotemporal shift-invariant recognition ability that is inherited from the time delay neural network (TDNN) and space displacement neural network (SDNN), where TDNN and SDNN are good at temporal and spatial shift-invariant recognition, respectively. In contrast to multilayer perceptron (MLP), TDNN, and SDNN, STDNN is constructed by vector-type nodes and matrix-type links such that the spatiotemporal information can be accurately represented in a neural network. Also evaluated herein is the performance of the proposed STDNN via two experiments. The moving Arabic numerals (MAN) experiment simulates the object's free movement in the space-time domain on image sequences. According to these results, STDNN possesses a good generalization ability with respect to the spatiotemporal shift-invariant recognition. In the lipreading experiment, STDNN recognizes the lip motions based on the inputs of real image sequences. This observation confirms that STDNN yields a better performance than the existing TDNN-based system, particularly in terms of the generalization ability. In addition to the lipreading application, the STDNN can be applied to other problems since no domain-dependent knowledge is used in the experiment.     

A statistical correlation technique and a neural network for the motion correspondence problem

A statistical correlation technique (SCT) and two variants of a neural network are presented to solve the motion correspondence problem. Solutions of the motion correspondence problem aim to maintain the identities of individuated elements as they move. In a pre-processing stage, two snapshots of a moving scene are convoluted with two-dimensional Gabor functions, which yields orientations and spatial frequencies of the snapshots at every position. In this paper these properties are used to extract, respectively, the attributes orientation, size and position of line segments. The SCT uses cross-correlations to find the correct translation components, angle of rotation and scaling factor. These parameters are then used in combination with the positions of the line segments to calculate the centre of motion. When all of these parameters are known, the new positions of the line segments from the first snapshot can be calculated and compared to the features in the second snapshot. This yields the solution of the motion correspondence problem. Since the SCT is an indirect way of solving the problem, the principles of the technique are implemented in interactive activation and competition neural networks. With boundary problems and noise these networks perform better than the SCT. They also have the advantage that at every stage of the calculations the best candidates for corresponding pairs of line segments are known.