Modeling visual attention via selective tuning

Using fuzzy logic to enhance stereo matching in multiresolution images

Stereo matching is an open problem in computer vision, for which local features are extracted to identify corresponding points in pairs of images. The results are heavily dependent on the initial steps. We apply image decomposition in multiresolution levels, for reducing the search space, computational time, and errors. We propose a solution to the problem of how deep (coarse) should the stereo measures start, trading between error minimization and time consumption, by starting stereo calculation at varying resolution levels, for each pixel, according to fuzzy decisions. Our heuristic enhances the overall execution time since it only employs deeper resolution levels when strictly necessary. It also reduces errors because it measures similarity between windows with enough details. We also compare our algorithm with a very fast multi-resolution approach, and one based on fuzzy logic. Our algorithm performs faster and/or better than all those approaches, becoming, thus, a good candidate for robotic vision applications. We also discuss the system architecture that efficiently implements our solution.

The time course of exogenous and endogenous control of covert attention

Studies of eye-movements and manual response have established that rapid overt selection is largely exogenously driven toward salient stimuli, whereas slower selection is largely endogenously driven to relevant objects. We use the N2pc, an event-related potential index of covert attention, to demonstrate that this time course reflects an underlying pattern in the deployment of covert attention. We find that shifts of attention that occur soon after the onset of a visual search array are directed toward salient, task-irrelevant visual stimuli and are associated with slow responses to the target. In contrast, slower shifts are target-directed and are associated with fast responses. The time course of exogenous and endogenous control provides a framework in which some inconsistent results in the capture literature might be reconciled; capture may occur when attention is rapidly deployed.

NLOOK: a computational attention model for robot vision

The computational models of visual attention, originally proposed as cognitive models of human attention, nowadays are being used as front-ends to some robotic vision systems, like automatic object recognition and landmark detection. However, these kinds of applications have different requirements from those originally proposed. More specifically, a robotic vision system must be relatively insensitive to 2D similarity transforms of the image, as in-plane translations, rotations, reflections and scales, and it should also select fixation points in scale as well as position. In this paper a new visual attention model, called NLOOK, is proposed. This model is validated through several experiments, which show that it is less sensitive to 2D similarity transforms than other two well known and publicly available visual attention models: NVT and SAFE. Besides, NLOOK can select more accurate fixations than other attention models, and it can select the scales of fixations, too. Thus, the proposed model is a good tool to be used in robot vision systems.

An Active Vision System for Detecting, Fixating and Manipulating Objects in the Real World

The ability to autonomously acquire new knowledge through interaction with the environment is an important research topic in the field of robotics. The knowledge can only be acquired if suitable perception— action capabilities are present: a robotic system has to be able to detect, attend to and manipulate objects in its surrounding. In this paper, we present the results of our long-term work in the area of vision-based sensing and control. The work on finding, attending, recognizing and manipulating objects in domestic environments is studied. We present a stereo-based vision system framework where aspects of top-down and bottom-up attention as well as foveated attention are put into focus and demonstrate how the system can be utilized for robotic object grasping.

Modeling Search for People in 900 Scenes: A combined source model of eye guidance

How predictable are human eye movements during search in real world scenes? We recorded 14 observers' eye movements as they performed a search task (person detection) in 912 outdoor scenes. Observers were highly consistent in the regions fixated during search, even when the target was absent from the scene. These eye movements were used to evaluate computational models of search guidance from three sources: saliency, target features, and scene context. Each of these models independently outperformed a cross-image control in predicting human fixations. Models that combined sources of guidance ultimately predicted 94% of human agreement, with the scene context component providing the most explanatory power. None of the models, however, could reach the precision and fidelity of an attentional map defined by human fixations. This work puts forth a benchmark for computational models of search in real world scenes. Further improvements in modeling should capture mechanisms underlying the selectivity of observer's fixations during search.

Discriminant saliency, the detection of suspicious coincidences, and applications to visual recognition

A discriminant formulation of top-down visual saliency, intrinsically connected to the recognition problem, is proposed. The new formulation is shown to be closely related to a number of classical principles for the organization of perceptual systems, including infomax, inference by detection of suspicious coincidences, classification with minimal uncertainty, and classification with minimum probability of error. The implementation of these principles with computational parsimony, by exploitation of the statistics of natural images, is investigated. It is shown that Barlow's principle of inference by the detection of suspicious coincidences enables computationally efficient saliency measures which are nearly optimal for classification. This principle is adopted for the solution of the two fundamental problems in discriminant saliency, feature selection and saliency detection. The resulting saliency detector is shown to have a number of interesting properties, and act effectively as a focus of attention mechanism for the selection of interest points according to their relevance for visual recognition. Experimental evidence shows that the selected points have good performance with respect to 1) the ability to localize objects embedded in significant amounts of clutter, 2) the ability to capture information relevant for image classification, and 3) the richness of the set of visual attributes that can be considered salient.

Spatio-temporal adaptation in the unsupervised development of networked visual neurons

There have been many computational models mimicking the visual cortex that are based on spatial adaptations of unsupervised neural networks. In this paper, we present a new model called neuronal cluster which includes spatial as well as temporal weights in its unified adaptation scheme. The "in-place" nature of the model is based on two biologically plausible learning rules, Hebbian rule and lateral inhibition. We present the mathematical demonstration that the temporal weights are derived from the delay in lateral inhibition. By training with the natural videos, this model can develop spatio-temporal features such as orientation selective cells, motion sensitive cells, and spatio-temporal complex cells. The unified nature of the adaptation scheme allows us to construct a multilayered and task-independent attention selection network which uses the same learning rule for edge, motion, and color detection, and we can use this network to engage in attention selection in both static and dynamic scenes.

Limited influence of perceptual organization on the precision of attentional control

The role of perceptual organization in the precision of attentional control was assessed in three experiments. Observers viewed circular arrays of disks that varied in density. One disk was cued, directing attention to that disk. A series of tones then indicated shifts of attention to the next disk that was of the same color (Experiments 1 and 2) or on the same depth plane (Experiment 3). In the homogeneous condition, all of the disks were the same color (Experiments 1 and 2) or on the same depth plane (Experiment 3). In the heterogeneous condition, the disks alternated in color (Experiments 1 and 2) or stereoscopically defined depth (Experiment 3). If the observers were able to limit attention to disks within a group, the effective density of the displays in the heterogeneous conditions should have been one half that in the homogeneous conditions. There was little evidence that the observers could do this, indicating a limited role of perceptual organization in the precision of attentional control.

Attention to faces modulates early face processing during low but not high face discriminability

In the present study, we investigated whether attention to faces results in sensory gain modulation. Participants were cued to attend either to faces or to scenes in superimposed face-scene images for which face discriminability was manipulated parametrically. The face-sensitive N170 event-related potential component was used as a measure of early face processing. Attention to faces modulated N170 amplitude, but only when faces were not highly discriminable. Additionally, directing attention to faces modulated later processing (~230-300 msec) for all discriminability levels. These results demonstrate that attention to faces can modulate perceptual processing of faces at multiple stages of processing, including early sensory levels. Critically, the early attentional benefit is present only when the "face signal" (i.e., the perceptual quality of the face) in the environment is suboptimal.

Bilateral field advantage in visual crowding

Thirty randomly oriented T's were presented in a circle around fixation at an eccentricity of 11 degrees such that each T was crowded by its neighbors. Two locations within the same hemifield (unilateral condition) or one location in each hemifield (bilateral condition) were precued for subsequent probing. Observers were then asked to report the orientation of a target T at one of these locations. A bilateral field advantage was found: target identification was better when the two precued targets were in different hemifields than when they were within the same hemifield. This bilateral advantage was absent when only targets were presented, without any distracters. Further controls showed that this advantage could not be attributed to differences between horizontal and vertical target alignments or to visual field anisotropies. A similar bilateral advantage has been reported for multiple object tracking (Alvarez, G. A., & Cavanagh, P. (2005). Independent resources for attentional tracking in the left and right visual fields. Psychological Science 16(8), 637-643) and other attentional tasks. Our results suggest that crowding also demonstrates separate attentional resources in the left and right hemifields. There was a cost to attending to two targets presented unilaterally over attending to a single target. However, this cost was reduced when the two crowded targets were in separate hemifields.

How to Compare Apples and Oranges: Infants' Object Identification Tested With Equally Salient Shape, Luminance and Color Changes

What kind of featural information do infants rely on when they are trying to recognize a previously seen object? The question of whether infants use certain features (e.g. shape or color) more than others (e. g. luminance), can only be studied legitimately if visual salience is controlled, since the magnitude of feature values - how noticeable and interesting they are - will affect results. We employed a novel methodology, 'Interdimensional Salience Mapping', that allowed us to quantify and calibrate salience changes along shape, luminance and color feature dimensions. We then compared 9-month-old infants' identification of objects, employing feature changes that were equally salient. These results show that infants more readily identify objects on the basis of color and shape than luminance. Additionally, we show that relative salience changes rapidly in infancy - in particular, we found significantly higher salience thresholds for color in younger (6.5-month-old) infants - but that individual differences within an age group are remarkably modest.

Attention reshapes center-surround receptive field structure in macaque cortical area MT

Directing spatial attention to a location inside the classical receptive field (cRF) of a neuron in macaque medial temporal area (MT) shifts the center of the cRF toward the attended location. Here we investigate the influence of spatial attention on the profile of the inhibitory surround present in many MT neurons. Two monkeys attended to the fixation point or to 1 of 2 random dot patterns (RDPs) placed inside or next to the cRF, whereas a third RDP (the probe) was briefly presented in quick succession across the cRF and surround. The probe presentation responses were used to compute a map of the excitatory receptive field and its inhibitory surround. Attention systematically reshapes the receptive field profile, independently shifting both center and surround toward the attended location. Furthermore, cRF size is changed as a function of relative distance to the attentional focus: attention inside the cRF shrinks it, whereas directing attention next to the cRF expands it. In addition, we find systematic changes in surround inhibition and cRF amplitude. This nonmultiplicative push-pull modulation of the receptive field's center-surround structure optimizes processing at and near the attentional focus to strengthen the representation of the attended stimulus while reducing influences from distractors.

A sparsity-enforcing method for learning face features

In this paper, we propose a new trainable system for selecting face features from over-complete dictionaries of image measurements. The starting point is an iterative thresholding algorithm which provides sparse solutions to linear systems of equations. Although the proposed methodology is quite general and could be applied to various image classification tasks, we focus here on the case study of face and eyes detection. For our initial representation, we adopt rectangular features in order to allow straightforward comparisons with existing techniques. For computational efficiency and memory saving requirements, instead of implementing the full optimization scheme on tenths of thousands of features, we propose a three-stage architecture which consists of finding first intermediate solutions to smaller size optimization problems, then merging the obtained results, and next applying further selection procedures. The devised system requires the solution of a number of independent problems, and, hence, the necessary computations could be implemented in parallel. Experimental results obtained on both benchmark and newly acquired face and eyes images indicate that our method is a serious competitor to other feature selection schemes recently popularized in computer vision for dealing with problems of real-time object detection. A major advantage of the proposed system is that it performs well even with relatively small training sets.

An Attentional Mechanism for Selecting Appropriate Actions Afforded by Graspable Objects

An object may afford a number of different actions. In this article, we show that an attentional mechanism inhibits competing motor programs that could elicit erroneous actions. Participants made a speeded key press to categorize the second of two successively presented door handles that were rotated at varying orientations relative to one another. Their responding hand was compatible or incompatible with the graspable part of the door handles (rightward or leftward facing). Compatible responses were faster than incompatible responses if the two handles shared an identical orientation, but they were slower if the two handles were aligned at slightly dissimilar orientations. Such suppressive surround effects are hallmarks of attentional processing in the visual domain, but they have never been observed behaviorally in the motor domain. This finding delineates a common mechanism involved in two of the most important functions of the brain: processing sensory data and preparing actions based on that information.

Visual attention on the sphere

Human visual system makes an extensive use of visual attention in order to select the most relevant information and speed-up the vision process. Inspired by visual attention, several computer models have been developed and many computer vision applications rely today on such models. However, the actual algorithms are not suitable to omnidirectional images, which contain a significant amount of geometrical distortion. In this paper, we present a novel computational approach that performs in spherical geometry and thus is suitable for omnidirectional images. Following one of the actual models of visual attention, the spherical saliency map is obtained by fusing together intensity, chromatic, and orientation spherical cue conspicuity maps that are themselves obtained through multiscale analysis on the sphere. Finally, the consecutive maxima in the spherical saliency map represent the spots of attention on the sphere. In the experimental part, the proposed method is then compared to the standard one using a synthetic image. Also, we provide examples of spots detection in real omnidirectional scenes which show its advantages. Finally, an experiment illustrates the homogeneity of the detected visual attention in omnidirectional images.

Modelling distractor devaluation (DD) and its neurophysiological correlates

A series of recent studies have shown that selective attention can influence the emotional value of both selected as well as ignored items. Specifically, ignored items (distractors) were consistently rated less positively in emotional evaluations, following attentional selection, relative to (typically) simultaneously presented items (targets). Furthermore, a known electrophysiological index of attentional selectivity (N2pc) was shown to correlate with the magnitude of the observed 'distractor devaluation' (DD). A neural model is presented here to account for these findings by means of a plausible mechanism linking attentional processes to emotional evaluations. This mechanism relies on the transformation of attentional inhibition of the distractor into a reduction of the value of that distractor. The model is successful in reproducing the existent behavioural results as well as the observed link between the magnitude of the attentional N2pc and the magnitude of DD. Moreover, the model proposes a series of testable hypotheses as well as specific predictions that call for further experimental investigation.

A (somewhat) new solution to the variable binding problem

To perform automatic, unconscious inference, the human brain must solve the binding problem by correctly grouping properties with objects. Temporal binding models like SHRUTI already suggest much of how this might be done in a connectionist and localist way by using temporal synchrony. We propose a set of alternatives to temporal synchrony mechanisms that instead use short signatures. This serves two functions: it allows us to explore an additional biologically plausible alternative, and it allows us to extend and improve the capabilities of these models. These extensions model the human ability to both perform unification and handle multiple instantiations of logical terms. To verify our model's feasibility, we simulate it with a computer system modeling simple, neuron-like computations.

Attention to stimulus features shifts spectral tuning of V4 neurons during natural vision

Previous neurophysiological studies suggest that attention can alter the baseline or gain of neurons in extrastriate visual areas but that it cannot change tuning. This suggests that neurons in visual cortex function as labeled lines whose meaning does not depend on task demands. To test this common assumption, we used a system identification approach to measure spatial frequency and orientation tuning in area V4 during two attentionally demanding visual search tasks, one that required fixation and one that allowed free viewing during search. We found that spatial attention modulates response baseline and gain but does not alter tuning, consistent with previous reports. In contrast, feature-based attention often shifts neuronal tuning. These tuning shifts are inconsistent with the labeled-line model and tend to enhance responses to stimulus features that distinguish the search target. Our data suggest that V4 neurons behave as matched filters that are dynamically tuned to optimize visual search.

Neuromorphic VLSI Models of Selective Attention: From Single Chip Vision Sensors to Multi-chip Systems

Biological organisms perform complex selective attention operations continuously and effortlessly. These operations allow them to quickly determine the motor actions to take in response to combinations of external stimuli and internal states, and to pay attention to subsets of sensory inputs suppressing non salient ones. Selective attention strategies are extremely effective in both natural and artificial systems which have to cope with large amounts of input data and have limited computational resources. One of the main computational primitives used to perform these selection operations is the Winner-Take-All (WTA) network. These types of networks are formed by arrays of coupled computational nodes that selectively amplify the strongest input signals, and suppress the weaker ones. Neuromorphic circuits are an optimal medium for constructing WTA networks and for implementing efficient hardware models of selective attention systems. In this paper we present an overview of selective attention systems based on neuromorphic WTA circuits ranging from single-chip vision sensors for selecting and tracking the position of salient features, to multi-chip systems implement saliency-map based models of selective attention.

The center-surround profile of the focus of attention arises from recurrent processing in visual cortex

We recently demonstrated with magnetoencephalographic recordings in human observers that the focus of attention in visual search has a spatial profile consisting of a center enhancement surrounded by a narrow zone of sensory attenuation. Here, we report new data from 2 experiments providing insights into the cortical processes that cause the surround attenuation. We show that surround suppression appears in search tasks that require spatial scrutiny, that is the precise binding of search-relevant features at the target's location but not in tasks that permit target discrimination without precise localization. Furthermore, we demonstrate that surround attenuation is linked with a stronger recurrent activity modulation in early visual cortex. Finally, we show that surround suppression appears with a delay (more than 175 ms) that is beyond the time course of the initial feedforward sweep of processing in the visual system. These observations together indicate that the suppressive surround is associated with recurrent processing and binding in the visual cortex.

Occipital network for figure/ground organization

To study the cortical mechanism of figure/ground categorization in the human brain, we employed fMRI and the temporal-asynchrony paradigm. This paradigm is able to eliminate any differential activation for local stimulus features, and thus to identify only global perceptual interactions. Strong segmentation of the image into different spatial configurations was generated solely from temporal asynchronies between zones of homogeneous dynamic noise. The figure/ground configuration was a single geometric figure enclosed in a larger surround region. In a control condition, the figure/ground organization was eliminated by segmenting the noise field into many identical temporal-asynchrony stripes. The manipulation of the type of perceptual organization triggered dramatic reorganization in the cortical activation pattern. The figure/ground configuration generated suppression of the ground representation (limited to early retinotopic visual cortex, V1 and V2) and strong activation in the motion complex hMT+/V5+; conversely, both responses were abolished when the figure/ground organization was eliminated. These results suggest that figure/ground processing is mediated by top-down suppression of the ground representation in the earliest visual areas V1/V2 through a signal arising in the motion complex. We propose a model of a recurrent cortical architecture incorporating suppressive feedback that operates in a topographic manner, forming a figure/ground categorization network distinct from that for "pure" scene segmentation and thus underlying the perceptual organization of dynamic scenes into cognitively relevant components.

Persuading visual attention through geometry

Artists, illustrators, photographers, and cinematographers have long used the principles of contrast and composition to guide visual attention. In this paper we introduce geometry modification as a tool to persuasively direct visual attention. We build upon recent advances in mesh saliency to develop techniques to alter geometry to elicit greater visual attention. Eye-tracking-based user studies show that our approach successfully guides user attention in a statistically significant manner. Our approach operates directly on geometry, and therefore, produces view-independent results that can be used with existing view-dependent techniques of visual persuasion.

Modeling the top-down influences on the lateral interactions in the visual cortex

Attention modulates the amount of excitatory and inhibitory lateral interactions in the visual cortex. A recurrent neural network is proposed to account for modulatory influence of top-down signals. In the model, two types of inhibitions are distinguished: dendritic and lateral inhibitions. Dendritic inhibition regulates the amount of impact that surrounding cells may exert on a target cell via the dendrites of excitatory neurons and the dendrites of subpopulation of inhibitory neurons mediating lateral inhibition. Attention increases the amount of dendritic inhibition and prevents contextual interactions, while it has no effect on the target cell when there is no surround input. Computer simulations showed that the proposed model is able to exhibit properties of attentional gating. In the condition of focused attention, neural activity in the presence of surrounding stimuli is restored to the level as when the target stimulus is presented alone. Moreover, the model is able to show contrast gain and response gain on the contrast sensitivity function depending on the strength of the dendritic inhibition.

On the relationship between flanker interference and localized attentional interference

The flanker interference (FI) effect suggests that visual attention operates like a mental spotlight, enhancing all stimuli within a selected region. In contrast, other data suggest difficulty dividing attention between objects near one another in the visual field, an effect termed localized attentional interference (LAI). The present experiment examined the relationship between these phenomena. Observers made speeded identity judgments of a colored target letter embedded among gray fillers. A response-compatible or -incompatible flanker of a non-target color appeared at varying distances from the target. Data gave evidence of LAI and spatially-graded FI, with mean RTs and flanker effects both decreasing with target-flanker separation. Both effects were reduced when target location was pre-cued and when the target was of higher salience than the flanker. Results suggest that the distribution of spatial attention modulates the strength of objects competing for selection, with this competition underlying both the FI and LAI effects.

Search goal tunes visual features optimally

How does a visual search goal modulate the activity of neurons encoding different visual features (e.g., color, direction of motion)? Previous research suggests that goal-driven attention enhances the gain of neurons representing the target's visual features. Here, we present mathematical and behavioral evidence that this strategy is suboptimal and that humans do not deploy it. We formally derive the optimal feature gain modulation theory, which combines information from both the target and distracting clutter to maximize the relative salience of the target. We qualitatively validate the theory against existing electrophysiological and psychophysical literature. A surprising prediction is that it is sometimes optimal to enhance nontarget features. We provide experimental evidence toward this through psychophysics experiments on human subjects, thus suggesting that humans deploy the optimal gain modulation strategy.

Using a filtering task to measure the spatial extent of selective attention

The spatial extent of attention was investigated by measuring sensitivity to stimuli at to-be-ignored locations. Observers detected a stimulus at a cued location (target), while ignoring otherwise identical stimuli at nearby locations (foils). Only an attentional cue distinguished target from foil. Several experiments varied the contrast and separation of targets and foils. Two theories of selection were compared: contrast gain and a version of attention switching called an all-or-none mixture model. Results included large effects of separation, rejection of the contrast gain model, and the measurement of the size and profile of the spatial extent of attention.

Gaze motion clustering in scan-path estimation

Visual attention is considered nowadays a paramount ability both in Cognitive Sciences and in Cognitive Vision to bridge the gap between perception and higher level reasoning functions, such as scene interpretation and decision making. Bottom-up gaze shifting is the main mechanism used by humans when exploring a scene without a specific task. In this paper we investigated which criteria allow for the generation of plausible fixation clusters by analysing experimental data of human subjects. We suggest that fixations should be grouped in cliques whose saliency can be assessed through an innovation factor encompassing bottom-up cues, proximity, direction and memory components.

Attentional costs in multiple-object tracking

Attentional demands of multiple-object tracking were demonstrated using a dual-task paradigm. Participants were asked to make speeded responses based on the pitch of a tone, while at the same time tracking four of eight identical dots. Tracking difficulty was manipulated either concurrent with or after the tone task. If increasing tracking difficulty increases attentional demands, its effect should be larger when it occurs concurrent with the tone. In Experiment 1, tracking difficulty was manipulated by having all dots briefly attract one another on some trials, causing a transient increase in dot proximity and speed. Results showed that increasing proximity and speed had a significantly larger effect when it occurred at the same time as the tone task. Experiments 2 and 3 showed that manipulating either proximity or speed independently was sufficient to produce this pattern of results. Experiment 4 manipulated object contrast, which affected tracking performance equally whether it occurred concurrent with or after the tone task. Overall, results support the view that the moment-to-moment tracking of multiple objects demands attention. Understanding what factors increase the attentional demands of tracking may help to explain why tracking is sometimes successful and at other times fails.

Attention and visual search

Selective Tuning (ST) presents a framework for modeling attention and in this work we show how it performs in covert visual search tasks by comparing its performance to human performance. Two implementations of ST have been developed. The Object Recognition Model recognizes and attends to simple objects formed by the conjunction of various features and the Motion Model recognizes and attends to motion patterns. The validity of the Object Recognition Model was first tested by successfully duplicating the results of Nagy and Sanchez. A second experiment was aimed at an evaluation of the model's performance against the observed continuum of search slopes for feature-conjunction searches of varying difficulty. The Motion Model was tested against two experiments dealing with searches in the visual motion domain. A simple odd-man-out search for counter-clockwise rotating octagons among identical clockwise rotating octagons produced linear increase in search time with the increase of set size. The second experiment was similar to one described by Thorton and Gilden. The results from both implementations agreed with the psychophysical data from the simulated experiments. We conclude that ST provides a valid explanatory mechanism for human covert visual search performance, an explanation going far beyond the conventional saliency map based explanations.

The role of context in object recognition

In the real world, objects never occur in isolation; they co-vary with other objects and particular environments, providing a rich source of contextual associations to be exploited by the visual system. A natural way of representing the context of an object is in terms of its relationship to other objects. Alternately, recent work has shown that a statistical summary of the scene provides a complementary and effective source of information for contextual inference, which enables humans to quickly guide their attention and eyes to regions of interest in natural scenes. A better understanding of how humans build such scene representations, and of the mechanisms of contextual analysis, will lead to a new generation of computer vision systems.

From physiological principles to computational models of the cortex

Understanding the brain goes through the assimilation of an increasing amount of biological data going from single cell recording to brain imaging studies and behavioral analysis. The description of cognition at these three levels provides us with a grid of analysis that can be exploited for the design of computational models. Beyond data related to specific tasks to be emulated by models, each of these levels also lays emphasis on principles of computation that must be obeyed to really implement biologically inspired computations. Similarly, the advantages of such a joint approach are twofold: computational models are a powerful tool to experiment brain theories and assess them on the implementation of realistic tasks, such as visual search tasks. They are also a way to explore and exploit an original formalism of asynchronous, distributed and adaptive computations with such precious properties as self-organization, emergence, robustness and more generally abilities to cope with an intelligent interaction with the world. In this article, we first discuss three levels at which a cortical circuit might be observed to provide a modeler with sufficient information to design a computational model and illustrate this principle with an application to the control of visual attention.

Observers are consistent when rating image conspicuity

Human perception of an image's conspicuity depends on the stimulus itself and the observer's semantic interpretation. We investigated the relative contribution of the former, sensory-driven, component. Participants viewed sequences of images from five different classes-fractals, overhead satellite imagery, grayscale and colored natural scenes, and magazine covers-and graded each numerically according to its perceived conspicuity. We found significant consistency in this rating within and between observers for all image categories. In a subsequent recognition memory test, performance was significantly above chance for all categories, with the weakest memory for satellite imagery, and reaching near ceiling for magazine covers. When repeating the experiment after one year, ratings remained consistent within each observer and category, despite the absence of explicit scene memory. Our findings suggest that the rating of image conspicuity is driven by image-immanent, sensory factors common to all observers.

Bio-inspired model of visual information encoding for localization: from the retina to the lateral geniculate nucleus

In this study, a bio-inspired approach for extracting efficient features prior to the recognition of scenes is proposed. It is highly inspired from the model of the mammals visual system. The retina contains many levels of neurons (bipolar, amacrine, horizontal and ganglion cells) accurately organized from cones and rods to the optic nerve up till the lateral geniculate nucleus (LGN) which is the main thalamic relay for inputs to the visual cortex. This structure probably eases other brain areas tasks in preprocessing the visual information. This paper is focusing on the study of these specific structures, relying on a bottom up approach to propose a comprehensive mathematical model of the low level image processing performed within the eye. The presented system takes into account the foveolar structure of the retina to produce a low-resolution representation of observed images by decomposing them into a local summation of elementary gaussian color histograms. This representation corresponds to the LGN biological organization. It has been thought that due to short timings, some very quick localization tasks involving particularly fast information processing pathways cannot be provided by the classical ones passing through higher level cortical areas. This work proposes a model of retinal coding and LGN-visual representation that we show provides reliable and sufficient early features for scenes recognition and localization. Experiments on real scenes using the developed model are presented showing the efficiency of the approach on localization.