Biological Vision Systems Research Articles

Response profiles of trajectory detectors

It has previously been demonstrated that detectors can be constructed that are sensitive to the 3D trajectory of a structure. The general approach is based upon a biological model for looming detectors proposed by K. Beverley and D. Regan (J. Physiol., vol.193, p.17-29, 1973). The computational model is based upon recent results in motion analysis and static stereopsis concerning the measurement of trajectory without prior form recognition. Some of the characteristics of these detectors are examined, and a number of experiments that show the responses of particular detectors to structure with different trajectories, disparities, and velocities are described. Some of the similarities and differences between the detectors presented and models for looming detectors present in biological vision systems are indicated.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Entropy reduction and decorrelation in visual coding by oriented neural receptive fields.

Image coding is optimally efficient if the coefficients of the representation are nonredundant, in the Shannon sense that their mutual information is zero. In that case, the code coefficients are uncorrelated and they form a statistically independent ensemble, so that the conditional probability of one coefficient's value, given any other coefficient's value, is the same as its unconditional probability: P(x/y) = P(x). In order for each coefficient to capture a unique property of the image that cannot be captured by any other coefficient, the expansion functions employed in the code must be linearly independent. In order for the code coefficients to have zero mutual information, the code primitives must be orthogonal so that their projections onto each other are always zero. In biological visual systems, although it is clear that some forms of efficiency (such as speed) are desirable, it is not obvious whether coding efficiency as measured by mutual information among the neurons is a factor which explains any of their properties. The center/surround receptive field profiles of neurons in the retina and geniculate are far from an orthogonal set, but a given neuron can still be regarded as a decorrelator of the incoming signal in the sense that it responds primarily to changes in the image (changes in space, time, chrominance, etc.) At the level of the brain's visual cortex, the introduction of the new variable of orientation selectivity can be regarded not only as a means for providing orientation labels for image structure, but also more basically as an effective decorrelator of the neural representation.(ABSTRACT TRUNCATED AT 250 WORDS)

Parallel integration of vision modules.

Computer algorithms have been developed for several early vision processes, such as edge detection, stereopsis, motion, texture, and color, that give separate cues to the distance from the viewer of three-dimensional surfaces, their shape, and their material properties. Not surprisingly, biological vision systems still greatly outperform computer vision programs. One of the keys to the reliability, flexibility, and robustness of biological vision systems is their ability to integrate several visual cues. A computational technique for integrating different visual cues has now been developed and implemented with encouraging results on a parallel supercomputer.

Pattern and motion vision without Laplacian zero crossings.

The zero crossings of band-limited signals are known to be rich in information. Recent models of information processing in biological visual systems have proposed that image structure is represented initially by such zero crossings, after the image has been bandpass filtered at multiple scales by neural receptive fields whose two-dimensional profiles resemble the Laplacian of a Gaussian. Because the resulting zero crossings generally correspond to physically meaningful structures such as edges and occlusion boundaries, and indeed under some conditions can exhaustively specify the original image, such representations have proven useful and efficient in a variety of machine vision problems. However, some simple information-processing operations that are apparent in human pattern and motion vision can be shown to be impossible in such representations, because the zero crossings in the bandpassed signals do not capture the necessary information (at any scale of analysis), because the information that they provide is misleading, or because there are no such crossings in the signals after delta 2 G sigma filtering at any scale. Examples are provided of simple visual signal-processing tasks (texture discrimination, motion perception, pattern detection) that human beings can perform effortlessly but that cannot be performed in the proposed multiscale delta 2 G sigma zero-crossings schemes. These perceptual capabilities offer evidence against this model of early image representation in human vision.

Infrared Transient Sensing

An experiment has been carried out to show that extrinsic silicon injection-mode infrared detectors can be used to simulate the transient response of neurons in biological vision systems. The neural response provides a means of high pass filtering in vision system focal plane parallel processing. The high pass filtering emphasizes transients in images. Possible IR imaging applications include surveillance, motion sensing, and tracking. The IR detector output pulses are sufficiently large that amplification is not required. This could facilitate parallel processing in or near the focal plane. The IR transient sensing approach is quantitatively analyzed in terms of a model for the detector and the associated circuitry. Excellent agreement is found between the model and the experimental data.

Computational studies of the extraction of visual spatial information from binocular and motion cues.

This paper reviews some of the contributions that work in computational vision has made to the study of biological vision systems. We concentrate on two areas where there has been strong interaction between computational and experimental studies: the use of binocular stereo to recover the distances to surfaces in space, and the recovery of the three-dimensional shape of objects from relative motion in the image. With regard to stereo, we consider models proposed for solving the stereo correspondence problem, focussing on the way in which physical properties of the world constrain possible methods of solution. We also show how critical observations regarding human stereo vision have helped to shape these models. With regard to the recovery of structure from motion, we focus on how the constraint of object rigidity has been used in computational models of this process.

Computing motion using analog and binary resistive networks

The authors describe recent developments in the theory of early vision that led from the formulation of the motion problem as an ill-posed one to its solution by minimizing certain 'cost' functions. These cost or energy functions can be mapped onto simple analog and digital resistive networks. The optical flow is computed by injecting currents into resistive networks and recording the resulting stationary voltage distribution at each node. The authors believe that these networks, which they implemented in complementary metal-oxide-semiconductor (CMOS) very-large-scale integrated (VLSI) circuits, represent plausible candidates for biological vision systems.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

The Emerging Paradigm of Computational Vision

Vision is the process by which structure in the world is inferred from images. But as such it is underconstrained; somehow additional constraints must be used to guide the inferences from two-dimensional (2-D) image data to three-dimensional (3-D) scene descriptions. The discovery of such constraints, and of techniques for using them, has led to the emergence of a new discipline called computational vision. It lies somewhere between the more classical attempts to solve engineering problems involving vision (as they arise, say, in robotics) and attempts to understand biological vision systems. Engineering applications suggest what should be solved; its scien­ tific counterparts suggest why; and biology suggests particular mechanisms that might be employed. But computational theories suggest how solutions can be obtained for different classes of visual tasks in the abstract, thereby filling the gap between particular problems and particular solutions. In this chapter I try to characterize the emergence of computational vision as its own discipline, contributing fundamental ideas to the study of both biological and engineered vision systems, rather than just supporting them with programming tools for testing existing ideas (cf Barlow et al 1972; Beck et al 1983; Braddick & Sleigh 1983). To illustrate the idea behind computational vision, consider an analogy with timekeeping. Mechanistic constraints come from clocks (the material from which they are composed, how they store energy, the mechanics of

A hierarchical cellular logic for pyramid computers

Hierarchical structure occurs in biological vision systems and there is good reason to incorporate it into a model of computation for processing binary images. A mathematical formalism is presented which can describe a wide variety of operations useful in image processing and graphics. The formalism allows for two kinds of simple transformations on the values (called pyramids) of a set of cells called a hierarchical domain: the first are binary operations on boolean values, and the second are neighborhood-matching operations. The implied model of computation is more structured than previously discussed pyramidal models, and is more readily realized in parallel hardware, while it remains sufficiently rich to provide efficient solutions to a wide variety of problems. The model has a simplicity which is due to the restricted nature of the operations and the implied synchronization across the hierarchical domain. A corresponding algebraic simplicity in the logic makes possible the concise representation of many cellular-data operations.

Application-oriented robotic vision – a review

SUMMARYComputer Vision is essentially concerned with emulating the process of seeing, naturally manifested in various higher biological systems,1–4 on a computational apparatus, and is consequently part of the Artificial Intelligence field within the sub-category of Machine perception. Seeing has to do with making sense of image data acquired through an optical system and subsequently dealt with at increasing levels of abstraction and association with known facts about the world. The spectrum of interest in Computer Vision ranges from attempting to answer basic questions concerning the functionality of biological vision systems, particularly human, at one end, all the way to enhancing the reliability, speed and cost effectiveness of specific industrial operations, particularly component inspection and vision driven robotic manipulation. The main bulk of interest is in the middle, where the quest for generality pushes interest towards biological vision systems with their demonstrated effectiveness in a wide range of environments, some hostile, whilst the need for economic viability and timeliness in relation to particular application pushes interest towards finding workable algorithms which function reliably at high speed on affordable apparatus.This paper is addressed, in somewhat tutorial style, at clarifying, by examples of work in the area, the issues surrounding application oriented robotic vision systems, their assumptions, strengths, weaknesses and degree of generality, and at the same time putting them in the context of the overall field of Computer Vision. In addition, the paper points to directions of development which promise to provide powerful industrial vision tools at an acceptable price.

The scope of motion research

This paper will attempt to briefly outline the broad scope of research endeavours that involve visual temporal change. It is acknowledged that several classes of researchers with differing backgrounds such as psychophysics and neurophysiology, as well as computer vision, artificial intelligence and computer graphics, are active in the area of motion research, and that their interaction may be of benefit. In general terms, these endeavours are those of quantification of intensity changes in image sequences, representation of visual temporal information, temporal reasoning and event description, motion generation, animation, control structures for coordinating the above into a unified computer system whether it be a vision system or an animation system, and studies of biological visual systems, both psychophysical and neurophysiological. Motion understanding will be briefly described in terms of its three main components: sensing motion, perception, interpretation and description of motion; and, motion generation. In an attempt to deal with the complexity of complete motion understanding systems, hypotheses and models will be drawn from the vast literature available in some of these different classes, although by necessity, this discussion will be short and limited.

The detection of intensity changes by computer and biological vision systems

This article describes the implementation of a theory for the detection of intensity changes, proposed by Marr and Hildreth ( Proc. R. Soc. London, Ser. B 207, 1980, 187–217). According to this theory, the image is first processed independently through a set of different size operators, whose shape is the Laplacian of a Gaussian, ▿ 2 G( x, y). The loci, along which the convolution outputs cross zero mark the positions of intensity changes at different resolutions. These zero-crossings can be described by their position, slope of the convolution output across zero, and two-dimensional orientation. The set of descriptions from different operator sizes forms the input for later visual processes, such as stereopsis and motion analysis. There are close parallels between this theory and the early processing of information by the human visual system.

Associative recall of images.

Orthogonal projection operations in a linear vector space are shown to have a close relation to the processes by which optimal associative recall of patterned information can be implemented. Two association schemes, the autoassociative mapping and the associative encoding, are introduced. The latter has a bearing on pattern recognition, especially in the recognition of an optical image from a small fragment of it. Analytical expressions for the quality of the recollections are derived, and computerized demonstrations of associative recall with 3024-element optical images are presented. Some preprocessing principles of the images are studied, whereby the two-dimensional Laplacian is found very effective. This finding may have some connection to lateral inhibition effects occurring in biological visual systems.

Mathematical Simulation of a Flexible-Lens- Based Imaging System

Standard optical imaging systems change their focal length by mechanical movement of lenses and lens groups along the optical axes. An alternative approach is realized in nature: many biological vision systems employ stationary flexible lenses enabling different lens power dependent of the surface shape. This paper presents an entirely Mathematica-based modelling and simulation of an innovative imaging system with a liquid-filled flexible lens design. Central issue is the investigation of the exact lens surface shape as a function of the internal liquid pressure. Therefore a new Mathematica package for Zerm/ce-Polynormals was realized. Zerm/ce-Polynomials allow surface data fits with an accuracy below 50 nm. Spline-Interpolation is used to parametrize Zerrafce-Coefncients over the entire pressure range applicable in the experiment. The lens surface function is implemented in the OPTICA package for exact modelling of the experimental setup. Raytracing and additional procedures for lens performance calculations have been developed. The modelling and simulation was realized on a Pentium Pro 200 MHz platform.