Recognition memory is thought to rely upon both recollection and familiarity. When people recall an episode from the past it is generally considered to reflect the memory process of recollection. Therefore, if people can successfully recall an item, they should be able to recognize it. However, in cued recall paradigms of memory research, participants sometimes correctly recall a studied target word in the presence of a strong semantic cue but then fail to recognize that word as actually having been studied. This paradox and underlying cognitive processes have been minimally studied by scientists, leaving this phenomenon poorly understood. Extant research has investigated some of the conditions necessary to produce these conditions but not the underlying neural correlates that drive them. The present study builds upon earlier studies using Electroencephalogram (EEG) to investigate the neural processes that underlie recognition failures of successfully recalled words. In the present experiment, participants studied words one at a time, and then later were asked to verbally recall these previously studied words as cued by their semantic associates. Following the participant’s verbal response, their recognition memory was tested for the recalled word. The current study aimed to use physiological measures (EEG) to investigate the explicit and implicit cognitive processes that may be involved in the recognition failure of recalled words. The data indicate that successfully recalled words that are recognized are driven by recollection at recall and a combination of recollection and familiarity at recognition, whereas successfully recalled words that are not recognized are instead driven by semantic priming at recall and at recognition, are driven by negative-going ERP effects reflecting implicit processes such as repetition fluency.

The Dunning-Kruger Effect is a metacognitive phenomenon in which individuals who perform poorly on a task believe they performed well, whereas individuals who performed very well believe their performance was only average. To date, this effect has only been investigated in the context of performance on mathematical, logical, or lexical tasks, but has yet to be explored for its generalizability in episodic memory task performance. We used a novel method to elicit the Dunning-Kruger Effect via a memory test of item and source recognition confidence. Participants studied 4 lists of words and were asked to make a simple decision about the words (source memory, i.e. Is it manmade? Is it alive?). They were later tested on their episodic memory and source memory for the words using a five-point recognition confidence scale, while electroencephalography (EEG) was recorded. After the test, participants were asked to estimate the percentile in which they performed compared to other students. Participants were separated into four quartiles based on their performance accuracy. Results showed that participants in all four groups estimated the same percentile for their performance. Participants in the bottom 25th percentile overestimated their percentile the most, while participants in the top 75th percentile slightly under-estimated their percentile, exhibiting the DKE and extending its phenomenon into studies of episodic memory. Groups were then re-categorized into participants that over-estimated, correctly estimated, and under-estimated their percentile estimate. Over-estimators responded significantly faster than under-estimators when estimating themselves as in the top percentile and they responded slower when evaluating themselves as in the bottom percentile. EEG first revealed generic scalp-wide differences within-subjects for all memory judgments as compared to all self-estimates of metacognition, indicating an effective sensitivity to task differences. More specific differences in late parietal sites were evident between high percentile estimates and low percentile estimates. Between-group differences were evident between over-estimators and under-estimators when collapsing across all Dunning-Kruger responses, which revealed a larger late parietal component (LPC) associated with recollection-based processing in under-estimators compared to those of over-estimators when assessing their memory judgements. These findings suggest that over- and under-estimators use differing cognitive strategies when assessing their performance and that under-estimators use less recollection when remembering episodic items, thereby revealing that episodic memory processes are playing a contributory role in the metacognitive judgments of illusory superiority that are characterized by the Dunning-Kruger Effect.

Modern systems produce a great amount of information and cues from which human operators must take action. On one hand, these complex systems can place a high demand on an operator's cognitive load, potentially overwhelming them and causing poor performance. On the other hand, some systems utilize extensive automation to accommodate their complexity; this can cause an operator to become complacent and inattentive, which again leads to deteriorated performance (Wilson, Russell, 2003a; Wilson, Russell, 2003b). An ideal human-machine interface would be one that optimizes the functional state of the operator, preventing overload while not permitting complacency, thus resulting in improved system performance. An operator's functional state (OFS) is the momentary ability of an operator to meet task demands with their cognitive resources. A high OFS indicates that an operator is vigilant and aware, with ample cognitive resources to achieve satisfactory performance. A low OFS, however, indicates a non-optimal cognitive load, either too much or too little, resulting in sub-par system performance (Wilson, Russell, 1999). With the ability to measure and detect changes in OFS in real-time, a closed-loop system between the operator and machine could optimize OFS through the dynamic allocation of tasks. For instance, if the system detects the operator is in cognitive overload, it can automate certain tasks allowing them to better focus on salient information. Conversely, if the system detects under-vigilance, it can allocate tasks back to the manual control of the operator. In essence, this system operates to "dynamically match task demands to [an] operator's momentary cognitive state", thereby achieving optimal OFS (Wilson, Russell, 2007). This concept is termed adaptive aiding and has been the subject of much research, with recent emphasis on accurately assessing OFS in real-time. OFS is commonly measured indirectly, like using overt performance metrics on tasks; if performance is declining, a low OFS is assumed. Another indirect measure is the subjective estimate of mental workload, where an operator narrates his/her perceived functional state while performing tasks (Wilson, Russell, 2007). Unfortunately, indirect measures of OFS are often infeasible in operational settings; performance metrics are difficult to construct for highly-automated complex systems, and subjective workload estimates are often inaccurate and intrusive (Wilson, Russell, 2007; Prinzel et al., 2000; Smith et al., 2001). OFS can be more directly measured via psychophysiological signals such as electroencephalogram (EEG) and electrooculography (EOG). Current research has demonstrated these signals' ability to respond to changing cognitive load and to measure OFS (Wilson, Fisher, 1991; Wilson, Fisher, 1995; Gevins et al., 1997; Gevins et al., 1998; Byrne, Parasuraman, 1996). Moreover, psychophysiological signals are continuously available and can be obtained in a non-intrusive manner, pre-requisite for their use in operational environments. The objective of this study is to advance schemes which detect change in OFS by monitoring psychophysiological signals in real-time. Reviews on similar methods can be found in, e.g., Wilson and Russell (2003a) and Wilson and Russell (2007). Many of these methods employ pattern recognition to classify mental workload into one of several discrete categories. For instance, given an experiment with easy, medium and hard tasks, and assuming the tasks induce varying degrees of mental workload on a subject, these methods classify which task is being performed for each epoch of psychophysiological data. The most common classifiers are artificial neural networks (ANN) and multivariate statistical techniques such as stepwise discriminant analysis (SWDA). ANNs have proved especially effective at classifying OFS as they account for the non-linear and higher order relationships often present in EEG/EOG data; they routinely achieve classification accuracy greater than 80%. However, the discrete output of these classification schemes is not conducive to real-time change detection. They accurately classify OFS, but they do not indicate when OFS has changed; the change points remain ambiguous and left to subjective interpretation. Thus, the present study introduces several online algorithms which objectively determine change in OFS via real-time psychophysiological signals. The following chapters describe the dataset evaluated, discuss the statistical properties of psychophysiological signals, and detail various algorithms which utilize these signals to detect real-time changes in OFS. The results of the algorithms are presented along with a discussion. Finally, the study is concluded with a comparison of each method and recommendations for future application.

Our research mainly consisted by two parts. First, apply p-norm error measure instead of 1-norm measure in a linear programming discrimination, which generates a linear hyperplane to classify two data sets. With this p-norm error measure, the errors generated by the classifier are not treated equally but rather biased. For 1, the bigger one error is, the more weight it obtains in the objective function. Second, investigation is conducted on a psychophysiological data set. Various methods are tested on this multi-dimensional time-series data set, from the linear programming method to the neural network method. With the help of DFT, The data is able to be transferred from the time domain to the frequency domain, in which the data set has interesting patterns

PhD / When modeling a system as complex as the brain, considerable simplifications are inevitable. The nature of these simplifications depends on the available experimental evidence, and the desired form of model predictions. A focus on the former often inspires models of networks of individual neurons, since properties of single cells are more easily measured than those of entire populations. However, if the goal is to describe the processes responsible for the electroencephalogram (EEG), such models can become unmanageable due to the large numbers of neurons involved. Mean-field models in which assemblies of neurons are represented by their average properties allow activity underlying the EEG to be captured in a tractable manner. The starting point of the results presented here is a recent physiologically-based mean-field model of the corticothalamic system, which includes populations of excitatory and inhibitory cortical neurons, and an excitatory population representing the thalamic relay nuclei, reciprocally connected with the cortex and the inhibitory thalamic reticular nucleus. The average firing rates of these populations depend nonlinearly on their membrane potentials, which are determined by afferent inputs after axonal propagation and dendritic and synaptic delays. It has been found that neuronal activity spreads in an approximately wavelike fashion across the cortex, which is modeled as a two-dimensional surface. On the basis of the literature, the EEG signal is assumed to be roughly proportional to the activity of cortical excitatory neurons, allowing physiological parameters to be extracted by inverse modeling of empirical EEG spectra. One objective of the present work is to characterize the statistical distributions of fitted model parameters in the healthy population. Variability of model parameters within and between individuals is assessed over time scales of minutes to more than a year, and compared with the variability of classical quantitative EEG (qEEG) parameters. These parameters are generally not normally distributed, and transformations toward the normal distribution are often used to facilitate statistical analysis. However, no single optimal transformation exists to render data distributions approximately normal. A uniformly applicable solution that not only yields data following the normal distribution as closely as possible, but also increases test-retest reliability, is described in Chapter 2. Specialized versions of this transformation have been known for some time in the statistical literature, but it has not previously found its way to the empirical sciences. Chapter 3 contains the study of intra-individual and inter-individual variability in model parameters, also providing a comparison of test-retest reliability with that of commonly used EEG spectral measures such as band powers and the frequency of the alpha peak. It is found that the combined model parameters provide a reliable characterization of an individual's EEG spectrum, where some parameters are more informative than others. Classical quantitative EEG measures are found to be somewhat more reproducible than model parameters. However, the latter have the advantage of providing direct connections with the underlying physiology. In addition, model parameters are complementary to classical measures in that they capture more information about spectral structure. Another conclusion from this work was that a few minutes of alert eyes-closed EEG already contain most of the individual variability likely to occur in this state on the scale of years. In Chapter 4, age trends in model parameters are investigated for a large sample of healthy subjects aged 6-86 years. Sex differences in parameter distributions and trends are considered in three age ranges, and related to the relevant literature. We also look at changes in inter-individual variance across age, and find that subjects are in many respects maximally different around adolescence. This study forms the basis for prospective comparisons with age trends in evoked response potentials (ERPs) and alpha peak morphology, besides providing a standard for the assessment of clinical data. It is the first study to report physiologically-based parameters for such a large sample of EEG data. The second main thrust of this work is toward incorporating the thalamocortical system and the basal ganglia in a unified framework. The basal ganglia are a group of gray matter structures reciprocally connected with the thalamus and cortex, both significantly influencing, and influenced by, their activity. Abnormalities in the basal ganglia are associated with various disorders, including schizophrenia, Huntington's disease, and Parkinson's disease. A model of the basal ganglia-thalamocortical system is presented in Chapter 5, and used to investigate changes in average firing rates often measured in parkinsonian patients and animal models of Parkinson's disease. Modeling results support the hypothesis that two pathways through the basal ganglia (the so-called direct and indirect pathways) are differentially affected by the dopamine depletion that is the hallmark of Parkinson's disease. However, alterations in other components of the system are also suggested by matching model predictions to experimental data. The dynamics of the model are explored in detail in Chapter 6. Electrophysiological aspects of Parkinson's disease include frequency reduction of the alpha peak, increased relative power at lower frequencies, and abnormal synchronized fluctuations in firing rates. It is shown that the same parameter variations that reproduce realistic changes in mean firing rates can also account for EEG frequency reduction by increasing the strength of the indirect pathway, which exerts an inhibitory effect on the cortex. Furthermore, even more strongly connected subcircuits in the indirect pathway can sustain limit cycle oscillations around 5 Hz, in accord with oscillations at this frequency often observed in tremulous patients. Additionally, oscillations around 20 Hz that are normally present in corticothalamic circuits can spread to the basal ganglia when both corticothalamic and indirect circuits have large gains. The model also accounts for changes in the responsiveness of the components of the basal ganglia-thalamocortical system, and increased synchronization upon dopamine depletion, which plausibly reflect the loss of specificity of neuronal signaling pathways in the parkinsonian basal ganglia. Thus, a parsimonious explanation is provided for many electrophysiological correlates of Parkinson's disease using a single set of parameter changes with respect to the healthy state. Overall, we conclude that mean-field models of brain electrophysiology possess a versatility that allows them to be usefully applied in a variety of scenarios. Such models allow information about underlying physiology to be extracted from the experimental EEG, complementing traditional measures that may be more statistically robust but do not provide a direct link with physiology. Furthermore, there is ample opportunity for future developments, extending the basic model to encompass different neuronal systems, connections, and mechanisms. The basal ganglia are an important addition, not only leading to unified explanations for many hitherto disparate phenomena, but also contributing to the validation of this form of modeling.

This research set out to develop a novel technique to decompose Electroencephalograph (EEG) signal into sets of constituent peaks in order to better describe the underlying nature of these signals. It began with the question; can a localised, single stimulation of sensory nervous tissue in the body be detected in the brain? Flash Visual Evoked Potential (VEP) tests were carried out on 3 participants by presenting a flash and recording the response in the occipital region of the cortex. By focussing on analysis techniques that retain a perspective across different domains � temporal (time), spectral (frequency/scale) and epoch (multiple events) � useful information was detected across multiple domains, which is not possible in single domain transform techniques. A comprehensive set of algorithms to decompose evoked potential data into sets of peaks was developed and tested using wavelet transform singularity detection methods. The set of extracted peaks then forms the basis for a subsequent clustering analysis which identifies sets of localised peaks that contribute the most towards the standard evoked response. The technique is quite novel as no closely similar work in research has been identified. New and valuable insights into the nature of an evoked potential signal have been identified. Although the number of stimuli required to calculate an Evoked Potential response has not been reduced, the amount of data contributing to this response has been effectively reduced by 75%. Therefore better examination of a small subset of the evoked potential data is possible. Furthermore, the response has been meaningfully decomposed into a small number (circa 20) of constituent peaksets that are defined in terms of the peak shape (time location, peak width and peak height) and number of peaks within the peak set. The question of why some evoked potential components appear more strongly than others is probed by this technique. Delineation between individual peak sizes and how often they occur is for the first time possible and this representation helps to provide an understanding of how particular evoked potentials components are made up. A major advantage of this techniques is the there are no pre-conditions, constraints or limitations. These techniques are highly relevant to all evoked potential modalities and other brain signal response applications � such as in brain-computer interface applications. Overall, a novel evoked potential technique has been described and tested. The results provide new insights into the nature of evoked potential peaks with potential application across various evoked potential modalities.

SUMMARY Introduction: There has recently been a great deal of interest in the role of synchronous high-frequency gamma oscillations in brain function. This interest has been motivated by an increasing body of evidence, that oscillations which are synchronous in phase across separated neuronal populations, may represent an important mechanism by which the brain binds or integrates spatially distributed processing activity which is related to the same object. Many models of schizophrenia suggest an impairment in the integration of brain processing, such as a loosening of associations, disconnection, defective multiple constraint organization, or cognitive dysmetria. This has led to recent speculation that abnormalities of high-frequency gamma synchronization may reflect a core dimension of the disturbance underlying this disorder. However, examination of the phase synchronization of gamma oscillations in patients with schizophrenia has never been previously undertaken. Method: In this thesis a new method of analysis of gamma synchrony was introduced, which enables the phase relationships of oscillations in a specific frequency band to be examined across multiple scalp sites as a function of time. This enabled, for the first time, the phase synchronization of gamma oscillations across widespread regions, to be studied in electrical brain activity measured at the scalp in humans. Gamma synchrony responses were studied in electroencephalographic (EEG) data acquired during a commonly employed conventional auditory oddball paradigm. The research consisted of two sets of experiments. In the first set of experiments, data from 100 normal subjects, consisting of 10 males and 10 females in each age decade from 20 to 70, was examined. These experiments were designed to characterize the gamma synchonizations that occurred in response to target and background stimuli and their functional significance in normal brain activity, and to exclude the possibility of these findings being due to electromyogram (EMG) or volume conduction artifact. The examination of functional significance involved the development of an additional new analysis technique. In the second set of experiments, data acquired from 35 patients with schizophrenia and 35 matched normal controls was analyzed. The purpose of these experiments was to determine whether patients showed disturbances of gamma synchrony compared to controls, and to establish the relationship of any such disturbances to medication levels, symptom profiles, duration of illness, and a range of psychophysiological variables. Results: In the 100 normals, responses to target stimuli were characterized by two bursts of synchronous gamma oscillations, an early (evoked) and a late (induced) synchronization, with different topographic distributions. Only the early gamma synchronization was seen in response to background stimuli. The main variable modulating the magnitude of these gamma synchronizations from epoch to epoch was pre-stimulus EEG theta (3-7 Hz) and delta (1-3 Hz) power. Early and late gamma synchrony were also associated with N1 and P3 ERP component amplitude across epochs. Across subjects, the early gamma synchronization was associated with shorter latency of the ERP components P2, N2 and P3, smaller amplitude of N1 and P2, and smaller pre-stimulus beta power. The control analyses showed that these gamma responses were specific to a narrow frequency range (37 to 41 Hz), and were not present in adjacent frequency bands. The responses were not generated by EMG contamination or volume conduction. In the 35 patients with schizophrenia, significant abnormalities of both the early and late synchronizations were observed compared to the 35 normal controls, with distinctive topographic characteristics. In general, early gamma synchrony was increased in patients compared to controls, and late gamma synchrony was decreased. These gamma synchrony disturbances were not related to medication level or the four summed symptom profile scores (positive, negative, general and total). They were, however, associated with duration of illness, becoming less severe the longer the patient had suffered from the disorder. The disordered gamma synchrony in patients was not secondary to abnormalities in other psychophysiological variables, but appeared to represent a primary disturbance. Discussion: The early synchronization may relate to the binding of object representations in early sensory processing, or, given that a constant inter-stimulus interval was employed, may be anticipatory and related to active memory. The late response is probably involved in binding in relation to activation of the internal contextual model involved in late expectancy/contextual processing (context updating or context closure) for target stimuli. The across epochs effects may relate to whether the focus of attention immediately prior to stimulus presentation is internal or is directed at the task. The across subjects effects suggest that a larger magnitude of the early gamma synchronization might indicate that the subject maintains a more stable and less ambiguous internal representation of the environment, that reduces the complexity of input and facilitates target/background discrimination and subsequent processing. The early gamma synchronization findings in patients with schizophrenia suggest that anticipatory processing involving active memory and forward-prediction of the environment is subject to over-binding or the formation of inappropriate associations. The late synchronization disturbances may reflect a fragmentation of contextual processing, and an inability to maintain contextual models of the environment intact over time. Conclusion: This research demonstrates the potential importance of integrative network activity as indexed by gamma phase synchrony in relation to normal cognition, and the possible broad relevance of such activity in psychiatric disorders. In particular, the application in this study to patients with schizophrenia showed that an impairment of brain integrative activity (missing links) might be a key feature of this illness.

Electroencephalography (EEG) is an electrical signal recorded from a person's scalp, and is used to monitor the neurological state of the patient. This thesis proposes a quantified continuity feature to aid preterm neonatal EEG analysis. The continuity of EEG signals for preterm infants refers to the variation of the EEG amplitude, and is affected by the conceptional age of the infants. Currently, the continuity of the signal is determined largely by visual examination of the raw EEG signal, or by using general guidelines on amplitude-integrated EEG (aEEG), which is a compressed plot of the estimated signal envelope. The proposed parametric feature embodies the statistical distribution parameters of the signal amplitudes. The signal is first segmented into pseudo-stationary segments using Generalized Likelihood Ratio (GLR). These segments are used to construct a vector of amplitude, the distribution of which can be modelled using a log-normal distribution. The mean and standard deviation of the log-normal distribution are used as the continuity feature. This feature is less prone to the effects of local transient activities than the aEEG. This investigation has demonstrated that the degree of continuity corresponds to the major axis of the feature distribution in the feature space, and the minor axis roughly corresponds to the age of the infants in healthy files. Principal component analysis was performed on the feature, with the first coefficient used as a continuity index and the second coefficient as a maturation index. In this research, classifiers were developed to use the continuity feature to produce a qualitative continuity label. It was found that using a linear discriminant analysis based classifier, labelled data can be used as training data to produce labels consistent across all recordings. It was also found that unsupervised classifiers can assist in identifying the intrinsic clusters occurring in the recordings. It was concluded that the proposed continuity feature can be used to aid further research in neonatal EEG analysis. Further work should focus on using the continuity information to provide a context for further feature extraction and analysis.

L'enjeu de la présente thèse est de proposer de nouvelles méthodes d'analyse des données d'imagerie cérébrale acquises en Imagerie par Résonance Magnétique fonctionnelle (IRMf). Elle s'est concentrée en particulier sur la compréhension des signaux temporels mesurés en IRMf et leur lien avec l'activité cérébrale. En effet, les variations du signal que l'on observe en IRMf sont dues à des changements de l'afflux du sang dans le cerveau et de l'oxygénation de ce sang. Ces changements sont liés à l'activité des neurones, et l'on nomme ce phénomène la réponse hémodynamique. Cette réponse hémodynamique fait l'objet d'un important effort de modélisation, de manière à mieux pouvoir interpréter les données d'IRMf. Et cette thèse contient des travaux liés à la fois à la modélisation pour elle-même, avec l'étude de certains détails des modèles hémodynamiques, et à la fois à l'utilisation de ces modèles pour l'analyse des données, avec en particulier l'analyse des données IRMf et la fusion entre des données d'IRMf et d'Electroencéphalographie (EEG). Ainsi, la première partie de la thèse est consacrée à l'utilisation de modèles hémodynamiques en IRMf. En effet, aujourd'hui, les méthodes standard d'analyse de données d'IRM fonctionnelle utilisent le Modèle Général Linéaire (GLM), qui suppose une relation linéaire entre l'activité des neurones, la réponse hémodynamique et les mesures IRMf. Nous montrons qu'il est aussi possible d'utiliser des modèles plus plausible du point de vue biologique, et éventuellement non-linéaires pour analyser les données. A la place de la régression linéaire utilisée habituellement, nous proposons une identification de modèle basées sur une minimisation d'énergie, et nous proposons d'adapter les tests de Fisher utilisés habituellement dans le cadre du GLM pour pouvoir réaliser dans le nouveau cadre la détection d'activations, le test d'hypothèses cognitives, ainsi que des comparaisons entre différents modèles. La seconde partie quant à elle est expérimentale: nous avons étudié les équations de différents modèles hémodynamiques grâce à des expérience d'Imagerie Optique chez le singe éveillé, dans le cadre d'une collaboration avec Ivo Vanzetta dans l'équipe "Dynamique de la perception visuelle et de l'action'' au CNRS Marseille. Nous nous sommes intéressés en particulier à la dynamique du flux sanguin, qui est de première importance car elle fait le lien entre les activités électriques et métaboliques et les changements du volume et de l'oxygénation du sang. Nous avons mis en évidence des aspects de la réponse hémodynamique qui ne sont pas prévus par les modèles actuels, tels qu'une non-linéarité de cette réponse du flux par rapport à l'intensité de la réponse électrique. Par ailleurs, dans le cadre de la même collaboration, nous avons conçu une méthode pour estimer la vitesse des globules rouges dans les vaisseaux sanguins filmés en Imagerie Optique, qui constitue une nouvelle technique de mesure de ce flux sanguin. Enfin, dans la troisième partie, nous avons étendu les méthodes présentées dans la première partie à l'analyse de données de modalités multiple, et en particulier, proposons une méthode pour estimer l'activité cérébrale à partir d'enregistrement simultanés en IRMf et en EEG. Cette méthode est validée sur des données synthétiques. Le présent synopsis résume les points importants de ces travaux: les objectifs, les méthodes, les conclusions et conséquences pour chaque chapitre. Nous avons également tenté d'en présenter une critique objective, en mentionnant à la fois ce qui constitue des contributions originales et les faiblesses restantes. Nous espérons que ce résumé permettra au lecteur de se repérer rapidement dans cette thèse, et de bien comprendre les relations entre ses différentes composantes.

Après un rapide rappel des principaux résultats de la psychologie et des neurosciences de la vision, illustrés par le schéma de Kosslyn, le parcours de six modèles computationnels de reconnaissance nous amène à discuter des principales alternatives élaborées pour décrire le traitement visuel – généralement compris comme complexe et récurrent. Le temps requis par ce traitement apparaît comme un critère crucial de décision sur son fonctionnement et d'affinement de notre compréhension.<br />Nous constatons que les données d'électrophysiologie disponibles ne permettent pas de disposer clairement de ce critère. Nous mettons alors en oeuvre une tâche expérimentale visant à mesurer le temps nécessaire au système visuel humain pour analyser des scènes naturelles contenant ou non un animal. Les résultats montrent que ce traitement peut être extrêmement rapide, d'une durée inférieure à 150ms. Cette première mesure est complétée par quatre expériences visant à mieux cerner cette contrainte temporelle, en variant les positions des images, leurs couleurs et la tâche. Cette vitesse du traitement visuel des scènes naturelles se montre particulière-ment robuste et constante : lors de présentations par hémichamps parafovéaux, lorsque l'attention n'est pas focalisée sur le lieu d'apparition du stimulus, et en l'absence de couleur comme indice de recon-naissance. Les résultats attenants montrent aussi que les catégorisations d'images contenant des formes simples et la détection de la présence de couleurs ne sont pas plus rapides. <br />La catégorisation "animal" semble d'autant plus résulter d'un mécanisme automatique que sa trace électrophysiologique est encore présente lorsqu'une autre tâche occupe les sujets.<br />Les bases cérébrales de la tâche ont été recherchées à l'aide de modèles dipolaires ainsi qu'avec la création d'un protocole événementiel d'imagerie cérébrale RMN analogue à celui mis en oeuvre en électrophysiologie. Nous montrons que cette tâche de<br />catégorisation implique de manière différentiée les aires visuelles extrastriées 19 et 31, le gyrus fusiforme et les cortex cingulaires postérieurs. Dans les aires visuelles, un effet de suppression d'activité neuronale lié à la présence d'une cible semble mettre en évidence le mécanisme de compétition postulée dans certains modèles.<br />Ces résultats plaident en faveur de mécanismes directs et rapides de la reconnaissance visuelle : traitement essentiellement ascendants (sans boucles) sans recentrage des stimuli latéralisés ; l'attention focalisée, la couleur et une forte acuité ne<br />sont pas nécessaires à la reconnaissance d'objets dans des scènes complexes.<br />La reconnaissance visuelle postulée comme mécanisme nécessitant des traitements<br />récurrents et des représentations complexes semble ainsi céder la place à de simples détections parallèles de traits visuels, en eux-mêmes suffisants à la représentation mentale des scènes. Dans ce cadre, la décision visuelle - le stimulus présent est adéquat à la tâche prévue – pourrait être l'extraction de ces représentations au moyen de l'inhibition des assemblées neuronales non sélectionnées.