Multidimensional gain control in image representation and processing in vision.

A generic model of automatic gain control (AGC) is proposed as a general framework for multidimensional automatic contrast sensitivity adjustment in vision, as well as in other sensory modalities. We show that a generic feedback AGC mechanism, incorporating a nonlinear synaptic interaction into the feedback loop of a neural network, can enhance and emphasize important image attributes, such as curvature, size, depth, convexity/concavity and more, similar to its role in the adjustment of photoreceptors and retinal network sensitivity over the extremely high dynamic range of environmental light intensities, while enhancing the contrast. We further propose that visual illusions, well established by psychophysical experiments, are a by-product of the multidimensional AGC. This hypothesis is supported by simulations implementing AGC, which reproduce psychophysical data regarding size contrast effects known as the Ebbinghaus illusion, and depth contrast effects. Processing of curvature by an AGC network illustrates that it is an important mechanism of image structure pre-emphasis, which thereby enhances saliency. It is argued that the generic neural network of AGC constitutes a universal, parsimonious, unified mechanism of neurobiological automatic contrast sensitivity control. This mechanism/model can account for a wide range of physiological and psychophysical phenomena, such as visual illusions and contour completion, in cases of occlusion, by a basic neural network. Likewise, and as important, biologically motivated AGC provides attractive new means for the development of intelligent computer vision systems.

Natural signal classification by neural cliques and phase-locked attractors.

Cortical neural networks are responsible for identification, recognition and classification of natural signals mediated by various sensory channels. These tasks are still too complex to be accomplished by state-of-the-art engineering systems. There is, therefore, a great deal of interest in the development of suitable biologically-motivated architectures which are based on a realistic model of generic neural ensembles. We present a computational architecture for classification of natural signals, such as physiological signals,based on the emergence of instant neural cliques and phase-locked attractors in liquid architectures. The emergence of instant neural cliques enables mapping of complex classes of signals onto specific spatio-temporal firing patterns. The convergence of neural cliques onto attractors, along phase-locked pathways, reveals a new type dynamic behavior of neural ensembles, which lends itself to simple discrete-output computational systems.

Fiber digestion and energy utilization of fat sand rats (Psammomys obesus) consuming the chenopod Anabasis articulata.

Most herbivorous rodents consume a variety of plants and, when available, select ones low in fiber content. In contrast, the fat sand rat (Psammomys obesus), a diurnal gerbillid rodent that is wholly herbivorous, is able to survive while consuming only the halophytic chenopod Anabasis articulata. We hypothesized that these gerbils are able to digest chenopods efficiently and that fiber digestion contributes substantially to their energy budget. Digestibility of total fibers (NDF), hemicellulose, and cellulose was 51.6%, 66.4%, and 44.0%, respectively. Energy derived from fiber digestion was 211.7 kJ kg(-0.75) d(-1) and resulted mostly from hemicellulose digestion. Average daily metabolic rate was 658.4 kJ kg(-0.75) d(-1), and basal metabolic rate (BMR) was determined to be close to 192.6 kJ kg(-0.75) d(-1). Therefore, fiber digestion provided 32% of maintenance requirements and 110% of BMR requirements, one of the highest values reported thus far for placental mammals. The efficiency of utilization of the chenopod was 0.29, a low value compared to other dietary items. We concluded that A. articulata fulfills all the energy and nutrient requirements of fat sand rats, even though energy yields and efficiency of utilization of its energy is low.

Multichannel wavelet-type decomposition of evoked potentials: model-based recognition of generator activity.

Scalp recording of electrical events allows the evaluation of human cerebral function, but contributions of the specific brain structures generating the recorded activity are ambiguous. This problem is ill-posed and cannot be solved without physiological constraints based on the spatio-temporal characteristics of the generators' activity. In our model-based analysis of evoked potentials for the purpose of generator activity detection, multichannel scalp-recorded signals are decomposed into a combination of wavelets, each of which can describe the neural mass coherent activity of cell assemblies. Elimination of contributions of specific generators and/or distributed background activity can produce physiologically motivated time-frequency filtering. The decomposition and filtering procedures are demonstrated by three examples; simulation of the surface manifestation of known intracranial generators; decomposition and reconstruction of auditory brainstem evoked potentials which reflect the differences among generators of these potentials; and cognitive components of evoked potentials which are diminished in the averaged recording but are clearly detected in single-trial signals.

The farthest point strategy for progressive image sampling.

A new method of farthest point strategy (FPS) for progressive image acquisition-an acquisition process that enables an approximation of the whole image at each sampling stage-is presented. Its main advantage is in retaining its uniformity with the increased density, providing efficient means for sparse image sampling and display. In contrast to previously presented stochastic approaches, the FPS guarantees the uniformity in a deterministic min-max sense. Within this uniformity criterion, the sampling points are irregularly spaced, exhibiting anti-aliasing properties comparable to those characteristic of the best available method (Poisson disk). A straightforward modification of the FPS yields an image-dependent adaptive sampling scheme. An efficient O(N log N) algorithm for both versions is introduced, and several applications of the FPS are discussed.

Interactions of monocular and cyclopean components and the role of depth in the Ebbinghaus illusion.

A study of size interactions of objects in three-dimensional space is reported. The canonical form of the Ebbinghaus illusion-test circles surrounded by large or small inducers-was used. Both monocularly visible (M) and purely cyclopean (C) objects were displayed stereoscopically to isolate the monocular and cyclopean components of the illusion. The results of two experiments indicate that: (i) depth plays a significant role when the test circles are cyclopean, but not when they are monocularly visible; and (ii) the size of C objects is affected equally by C and M inducers, but the size of M objects is affected much more strongly by M than by C inducers. In conclusion, possible explanations are offered for the main trends in the data, the most interesting of which is that cyclopean tests seem to be interacting only with the cyclopean component of monocularly visible inducers.

The design of two-dimensional gradient estimators based on one-dimensional operators.

Two-dimensional (2-D) gradient estimators are some of the most useful tools in image processing. A computational procedure for the extension of one-dimensional (1-D) gradient estimators to two dimensions (2-D) is presented. The procedure is equivalent to the surface fitting method. It is, however, simpler in design, as the design is 1-D rather than 2-D. Higher order derivative estimators can also be constructed by the same procedure.

Visual assessment of variable-resolution imagery.

A technique is described for producing variable-resolution images whose spatial detail decreases as a function of distance from their centers. These images can be matched in some sense to the normal spatial inhomogeneities of the human visual system, as well as to various abnormalities in spatial discrimination. A set of images was generated with a series of linear distortion functions whose low-pass characteristics differed at both the center and the periphery of the image as well as across the image. A forced-choice procedure was used to determine which test images were indistinguishable from unprocessed versions of themselves. Certain of the threshold distortion functions are compared with eccentricity scaling functions that have been used by others to characterize various aspects of peripheral vision. Finally, the concept of locally band-limited spaces is discussed, and an efficient sampling technique based on the concept is described. This technique can be used to generate an image that, under certain conditions, is visually equivalent to an otherwise identical image containing significantly more information.

Spatio-temporal multiple source localization by wavelet-type decomposition of evoked potentials.

Scalp recording of electrical events allows evaluation of human cerebral function, but contributions of the specific brain structures generating the recorded activity are ambiguous. This problem is ill-posed and cannot be solved without auxiliary physiological knowledge about the spatio-temporal characteristics of the generators' activity. In our source localization by model-based wavelet-type decomposition, scalp recorded signals are decomposed into a combination of wavelets, each of which may describe the coherent activity of a population of neurons. We chose the Hermite functions (derived from the Gaussian function to form mono-, bi- and triphasic wave forms) as the mathematical model to describe the temporal pattern of mass neural activity. For each wavelet we solve the inverse problem for two symmetrically positioned and oriented dipoles, one of which attains zero magnitude when a single source is more suitable. We use the wavelet to model the temporal activity pattern of the symmetrical dipoles. By this we reduce the dimension of inverse problem and find a plausible solution. Once the number and the initial parameters of the sources are given, we can apply multiple source localization to correct the solution for generators with overlapping activities. Application of the procedure to subcortical and cortical components of somatosensory evoked potentials demonstrates its feasibility.

Source estimation of auditory brainstem evoked potentials: comparison of 3CLT and dipole localization.

The generators of auditory brainstem evoked potentials (ABEPs) are generally agreed to be located between the auditory nerve and upper pons. Thus, they are all located within a few cm from the center of the head. Three-channel Lissajous' trajectory (3CLT) provides the amplitude and orientation of a centrally located equivalent dipole of surface recorded activity. Volume conductor theory predicts decreased spatial resolution of source estimation the deeper the source. In this study we compared source estimates obtained with 3CLT, using three orthogonal differential channels, with those obtained with two other source estimation methods: i) setting the generators at their known anatomical coordinates and calculating orientation and magnitude of the source (dipole localization method--DLM); ii) estimation of all source parameters, including the number of sources by wavelet-type decomposition, without assumptions on the location of the sources (multiple source estimate--MSE). 3CLT, DLM and MSE all converged on magnitudes and orientations that were not significantly different from each other, and locations that were within a few cm of each other. In conclusion, although 3CLT can only estimate a single, centrally located equivalent dipole, in the specific case of ABEPs, it provides the same information available from the more demanding source estimate methods. In addition to the considerable saving in recording channels, 3CLT is reference-independent and thus avoids ambiguities resulting from the choice of reference.

Preattentive equivalence of multicomponent Gabor textures in the central and peripheral visual field.

Similarity ratings were obtained to determine the minimum number of Gabor components that would produce a comparison texture that appeared preattentively similar to a 64-component standard texture. All textures were chosen to be both specifiable by a relatively small number of localized spectral components and sufficiently complex to approximate natural textures. The number of component orientations in the set of comparison textures was found to be a particularly important determinant of texture discrimination in that its effect on rated similarity was largely independent of the total number of components making up the texture. Textures were also presented at 0.75 degree and 20 degrees eccentricity, with the latter magnified by a factor of either 2 or 4. The overall similarity rating did not change with either magnification, whereas the critical number of orientations, defined as the number of orientations above which rated similarity was constant, did change for the higher magnification. The latter finding is consistent with the proposition that higher-order discriminations are mediated by higher cortical areas that integrate information across the visual field. Finally, the phase-bandwidth of a set of coherent textures was also varied in order to determine whether more explicit differences in the spatial structure of stimuli might affect rated similarity. In contrast to the results for component orientation, the ratings, obtained at 0.75 degree and 20 degrees, were different even when the phase-bandwidth stimuli were magnified by a factor of 4.

Nonuniform image representation in area-of-interest systems.

This paper is concerned with image representation by data distributed nonuniformly and in particular with a representation scheme suitable for "area-of-interest" imaging. The class of signals under consideration, whose information density varies with position, can be represented according to a nonuniform sampling scheme. Position-varying projection operators are presented as simple low-pass filtering operations in a Fourier-like domain. Sequential projections are used for pyramidal representation of nonuniformly sampled images. It is shown that irregular random sampling, prevents, under certain mild restrictions, aliasing effects.

Applications of limited-extent waves: an introduction.

This feature issue of Applied Optics is devoted to the theoretical and the experimental aspects of Gabor and wavelet transforms and to their optical implementations and applications. Subjects related to the Gabor scheme and wavelets have evolved under the influence of ideas originating primarily in physics and engineering. The powerful mathematical formalism and techniques that have been recently introduced, have, however, stimulated much greater interest among scientists of various disciplines and backgrounds. The purpose of this feature issue of Applied Optics is to further stimulate interest in this topic in the community of scientists working on optical techniques, in which it is expected that important applications of these concepts and techniques have yet to be developed.

The role of stimulus structure in spatial hyperacuity.

In spatial hyperacuity the subjects discriminate a stimulus feature relative to a reference, with an accuracy significantly better than the grain of the retinal mosaic. We show that the normalized thresholds have a dichotomous behavior; they are either insensitive to the spatial parameter in the experiment or increase very steeply with it. This behavior is explained by the involvement in the processing of pixel (receptor) accuracy information about the structure of the stimulus. A computational model employing optimal filtering reproduces the experimental data and suggests that processing of spatial hyperacuity tasks in the human visual system is optimal.

Localized texture processing in vision: analysis and synthesis in the Gaborian space.

Recent studies of cortical simple cell function suggest that the primitives of image representation in vision have a wavelet form similar to Gabor elementary functions (EF's). It is shown that textures and fully-textured images can be practically decomposed into, and synthesized from, a finite set of EF's. Textured-images can be synthesized from a set of EF's using image coefficient library. Alternatively, texturing of contoured (cartoon-like) images is analogous to adding chromaticity information to contoured images. A method for texture discrimination and image segmentation using local features based on the Gabor approach is introduced. Features related to the EF's parameters provide efficient means for texture discrimination and classification. This method is invariant under rotation and translation. The performance of the classification appears to be robust with respect to noisy conditions. The results show an insensitivity of the discrimination to relatively high noise levels, comparable to the performances of the human observer.

Preferences and asymmetries in saccadic responses to delayed bihemifield stimuli.

Saccaadic response preferences and latencies were measured using dual targets presented at 5, 10 and 15 deg along the horizontal meridian in either one or both visual hemifields. In the unihemifield condition, subjects exhibited a strong preference (about 95%) in favor of the target presented closer to the original fixation point. Further, the presence of the second target did not increase the response latency to the preferred target. In the bihemifield condition using spatially symmetric targets, directional preference varied across subjects from 57 to 100% (mean = 83%). There was also a significant increase in response latency as compared with responses to either single targets or dual targets presented in one hemifield. This latency increase was observed for responses in both the preferred and nonpreferred directions and was taken as evidence of mutual interaction between the two hemispheres. The response preferences could be eliminated by delaying the onset of the preferred target by about 100 msec in the unihemifield condition and by about 50 msec in the bihemifield condition. Further, the interactions between the effects of the targets on response time in the bihemifield condition were found to be asymmetrical in that a target presented in the preferred hemifield has a greater effect than a target in the nonpreferred hemifield on responses to targets presented in the counterhemifield. The results suggest an asymmetry in hemispheric interaction which has not previously been demonstrated in the visualoculomotor system.

Changes in the oscillatory potentials of the electroretinogram in glaucoma.

Dark adapted and progressively light adapted electroretinograms (ERGs) were recorded from 34 normals and from 45 glaucoma patients. To enhance the oscillatory potentials (OP) the ERGs were highpass-filtered. The OP were characterized by two indices: their root-mean-square value and the inter-flash interval for which the maximal amplitude was obtained. In most of the glaucomatous OP, 50 of 81 eyes, both indices were abnormal, in 22 one parameter was abnormal and only in 9 were both parameters normal. In 13 of the 14 OP recorded from the opposite ('normal') eye of patients with unilateral glaucoma either one or both of the indices were abnormal. These findings indicate that the damage to the retina in glaucoma may extend more distally than the ganglion cell layer and that subclinical changes might be revealed by OP changes.

Interhemispheric coordination is compromised in subjects with developmental dyslexia.

Developmental dyslexics (DDs) and good readers (GRs) were tested on measures of interhemispheric coordination. All subjects (ages 16 to 47) demonstrated normal oculomotor control and visual acuity, prior to testing. Subjects were instructed to track three different point-light source patterns (single stimulus in one hemifield, dual stimuli in one hemifield and a pair of simultaneous, symmetric, bihemifield stimuli [SSBS]), presented in random sequence and arrayed horizontally at +/- 5, +/- 10, and +/- 15 degrees of eccentricity. Tested with unihemifield stimuli, all subjects showed normal saccadic latencies and trajectories. In response to SSBS, all GRs had pronounced directional preference, choosing largely to track one side over the other. In contrast, DDs showed reduced laterality bias (p less than .025). DDs exhibited also significantly longer response latencies to SSBS than to unihemifield stimulation (p less than .01) and differed significantly from GRs (p less than .05).

Importance of global information in vernier acuity.

Noisy vernier configurations were generated by normally distributing the lateral displacements of dot elements forming vertical dotted lines. Observations with noisy configurations were compared with thresholds obtained with noise-free lines. In configurations with gap, we have found, through independent perturbation of the end regions adjacent to the gap, that the entire length of the line was actively processed. In an experiment with noisy abutting lines, the threshold as a function of line length exhibited only a slight upward shift relative to the curve obtained with noise-free configurations. Since it is not possible to interpret all the results through a comparison of local positional information extracted from the stimulus configuration, we conclude that global information is also involved in the processing.

Bihemifield visual stimulation reveals reduced lateral bias in dyslexia.

Orton (1936) observed that dyslexic readers display not only obvious linguistic processing errors, but also diminished lateralized specialization of other cerebral hemispheric functions. To explore his "intergrading" hypothesis, six developmental dyslexics (DDs) and a group of good readers (GRs) were tested on measures of interhemispheric coordination. All subjects (ages 16 to 47) demonstrated normal oculomotor control and visual acuity prior to testing. Subjects were instructed to track three different point-light source patterns, (single stimulus in one hemifield, dual stimuli in one hemifield and a pair of simultaneous, symmetric, bihemifield stimuli [SSBS]), presented in random sequence and arrayed horizontally at ±5, ±10, and ±15 degrees eccentricity. Tested with unihemifield stimuli, all subjects showed normal saccadic latencies and trajectories. In response to SSBS, all GRs showed pronounced directional preference, choosing largely to track one side over the other. DDs showed reduced laterality bias (p<.025). DDs showed significantly longer response latencies to SSBS than to unihemifield stimulation (p<.01) and differed significantly from GRs (p<.05).