The human visual system has learned to assume that light originates from above, most likely because of the persistent natural overhead light source – the sun (Ramachandran, 1988). Asymmetries of perception in neurologically normal individuals, like assuming light is coming from above, in part result from efficiency measures of the visual system. Not only is light assumed to come from above, but light from above and to the left has been found to decrease reaction times in target finding as well as increase aesthetic preference (Sun & Perona, 1998; Smith & Elias, 2013). The underlying cause of the bias towards upper-left lighting is debated, and may have a relationship with another peculiar phenomenon in neurologically normal individuals where greater attention is paid to leftward space, called pseudoneglect (Bowers & Heilman, 1980). Alternatively, an explanation suggesting that directional reading influences lighting preferences has been proposed, as Smith and Elias (2013) found native right-to-left readers to be significantly different from leftward biased left-to-right readers. The current set of experiments used eye-tracking and a target finding paradigm to assess differences between left-to-right and right-to-left readers. Manipulating the position of the light illuminating a field of spheres generated targets, creating either 1 convex bubble among 15 concave depressions, or vice-versa. Results from these studies are mixed, and highlight differences between both upper and lower and lateral visual space. Light originating from above facilitated shorter average duration times for both groups, whereas left-to-right readers tended to prefer light from the upper-left, while right-to-left readers preferred light from the upper-right. No one target location in the array facilitated shorter average duration times for right-to-left readers, although left-to-right readers tended to exhibit shorter durations when identifying targets in the upper-left quadrant. Participants spent the greatest amount of time examining the upper quadrants of the array, tending to focus more on the side of the image that their native reading direction begins on. The influence of directional reading on light source perception, and the potential problems of using exclusively Western participant samples are discussed.

Several decades of physiology, imaging and psychophysics research on attention has generated an enormous amount of data describing myriad forms of attentional effects. A similar breadth of theoretical models have been proposed that attempt to explain these effects in varying amounts of detail. However, there remains a need for neurally detailed mechanistic models of attention that connect more directly with various kinds of experimental data -- behavioural, psychophysical, neurophysiological, and neuroanatomical -- and that provide experimentally testable predictions. Research has been conducted that aims to identify neurally consistent principles that underlie selective attentional processing in cortex. The research specifically focuses on describing the functional mechanisms of attentional routing in a large-scale hierarchical model, and demonstrating the biological plausibility of the model by presenting a spiking neuron implementation that can account for a variety of attentional effects. The thesis begins by discussing several significant physiological effects of attention, and prominent brain areas involved in selective attention, which provide strong constraints for developing a model of attentional processing in cortex. Several prominent models of attention are then discussed, from which a set of common limitations in existing models is assembled that need to be addressed by the proposed model. One central limitation is that, for many existing models, it remains to be demonstrated that their computations can be plausibly performed in spiking neurons. Further, few models address attentional effects for more than a single neuron or single cortical area. And finally, few are able to account for different forms of attentional modulation in a single detailed model. These and other limitations are addressed by the Attentional Routing Circuit (ARC) proposed in this thesis. The presentation of the ARC begins with the proposal of a high-level mathematical model for selective routing in the visual hierarchy. The mathematical model is used to demonstrate that the suggested mechanisms allow for scale- and position-invariant representations of attended stimuli to be formed, and provides a functional context for interpreting detailed physiological effects. To evaluate the model's biological plausibility, the Neural Engineering Framework (NEF) is used to implement the ARC as a detailed spiking neuron model. Simulation results are then presented which demonstrate that selective routing can be performed efficiently in spiking neurons in a way that is consistent with the mathematical model. The neural circuitry for computing and applying attentional control signals in the ARC is then mapped on to neural populations in specific cortical laminae using known anatomical interlaminar and interareal connections to support the plausibility of its cortical implementation. The model is then tested for its ability to account for several forms of attentional modulation that have been reported in neurophysiological experiments. Three experiments of attention in macaque are simulated using the ARC, and for each of these experiments, the model is shown to be quantitatively consistent with measured data. Specifically, a study by Womelsdorf et al. (2008) demonstrates that spatial shifts of attention result in a shifting and shrinking of receptive fields depending on the target's position. An experiment by Treue and Martinez-Trujllo (1999) reports that attentional shifts between receptive field stimuli produce a multiplicative scaling of responses, but do not affect the neural tuning sensitivity. Finally, a study by Lee and Maunsell (2010) demonstrates that attentional shifts result in a multiplicative scaling of neural contrast-response functions that is consistent with a response-gain effect. The model accounts for each of these experimentally observed attentional effects using a single mechanism for selectively processing attended stimuli. In conclusion, it is suggested that the ARC is distinguished from previous models by providing a unifying interpretation of attentional effects at the level of single cells, neural populations, cortical areas, and over the bulk of the visual hierarchy. As well, there are several advantages of the ARC over previous models, including: (1) scalability to larger implementations without affecting the model's principles; (2) a significant increase in biological plausibility; (3) the ability to account for experimental results at multiple levels of analysis; (4) a detailed description of the model's anatomical substrate; (5) the ability to perform selective routing while preserving biological detail; and (6) generating a variety of experimentally testable predictions.

This thesis has presented a computational model for the combination of bottom-up and top-down attentional mechanisms. Furthermore, the use for this model has been demonstrated in a variety of applications of machine and robotic vision. We have observed that an attentional mechanism is imperative in any active vision system, machine as well as biological, since it not only reduces the amount of information that needs to be further processed (for say recognition, action), but also by only processing the attended image regions, such tasks become more robust to large amounts of clutter and noise in the visual field. Using various feature channels such as color, orientation, texture, depth and symmetry, as input, the presented model is able with a pre-trained artificial neural network to modulate a saliency map for a particular top-down goal, e.g. visual search for a target object. More specifically it dynamically combines the unmodulated bottom-up saliency with the modulated top-down saliency, by means of a biologically and psychophysically motivated temporal differential equation. This way the system is for instance able to detect important bottom-up cues, even while in visual search mode (top-down) for a particular object. All the computational steps for yielding the final attentional map, that ranks regions in images according to their importance for the system, are shown to be biologically plausible. It has also been demonstrated that the presented attentional model facilitates tasks other than visual search. For instance, using the covert attentional peaks that the model returns, we can improve scene understanding and segmentation through clustering or scattering of the 2D/3D components of the scene, depending on the configuration of these attentional peaks and their relations to other attributes of the scene. More specifically this is performed by means of entropy optimization of the scence under varying cluster-configurations, i.e. different groupings of the various components of the scene. Qualitative experiments demonstrated the use of this attentional model on a robotic humanoid platform and in a real-time manner control the overt attention of the robot by specifying the saccadic movements of the robot head. These experiments also exposed another highly important aspect of the model; its temporal variability, as opposed to many other attentional (saliency) models that exclusively deal with static images. Here the dynamic aspects of the attentional mechanism proved to allow for a temporally varying trade-off between top-down and bottom-up influences depending on changes in the environment of the robot. The thesis has also lay forward systematic and quantitative large scale experiments on the actual benefits and uses of this kind of attentional model. To this end a simulated 2D environment was implemented, where the system could not “see” the entire environment and needed to perform overt shifts of attention (a simulated saccade) in order to perfom a visual search task for a pre-defined sought object. This allowed for a simple and rapid substitution of the core attentional-model of the system with comparative computational models designed by other researchers. Nine such contending models were tested and compared with the presented model, in a quantitative manner. Given certain asumptions these experiments showed that the attentional model presented in this work outperforms the other models in simple visualsearch tasks. / QC 20111228

By visual attention process biological and machine vision systems are able to select the most relevant regions from a scene. The relevancy process is achieved either by top-down factors, driven by task, or bottom-up factors, the visual saliency, which distinguish a scene region that are different from its surrounding. During the past 20 years numerous research efforts have aimed to model bottom-up visual saliency with many successful applications in computer vision and robotics.In this thesis we have performed a comparison between a state-of-the-art saliency model and subjective test (human eye tracking) using different evaluation methods over three generated dataset of synthetic patterns and natural images. Our results showed that the objective model is partially valid and highly center-biased.By using empirical data obtained from subjective experiments we propose a special function, the Probability of Characteristic Radially Dependency Function, to model the lateral distribution of visual attention process.

L’attention est le processus qui consiste à filtrer les informations utiles à l’activité, de celles qui lui sont inutiles. La réalité augmentée (RA) de surlignage guide ce processus de sélection de l’information, en mettant en valeur certains éléments par rapport à d’autres. Telle qu’envisagée actuellement pour la conduite automobile, la RA de surlignage met en évidence des éléments liés à l’activité générale de conduite (e.g. panneaux en cas de mauvaise visibilité, direction à emprunter), mais indépendamment des manœuvres. Or, la littérature sur l’attention visuelle en activité nous montre que les parcours oculaires sont très spécifiques aux buts et sous-buts immédiats. Une RA qui ne respecte pas cette priorisation « naturelle » du traitement de l’information risque donc de perturber la prise d’informations.Le premier objectif de cette recherche est de déterminer dans quelle mesure l’allocation de l’attention visuelle en conduite automobile se concentre sur les informations liées à la manœuvre. Le second objectif est d’étudier l’impact de la RA sur cette allocation de l’attention.Nous avons mis en place trois expérimentations dans lesquelles les participants visualisaient des scènes statiques et dynamique de conduite automobile, et devaient décider s’ils pouvaient réaliser une manœuvre. Nous avons analysé les variations d’allocation de l’attention visuelle selon les manœuvres notifiées et les conditions de RA à l’aide d’enregistrements oculométriques. Nos résultats montrent que l’attention visuelle est fortement allouée aux indices permettant la prise de décision, mais qu’elle ne néglige pas les indices permettant la compréhension générale de la scène. La RA optimise l’attention visuelle lorsqu’elle met en évidence des indices liés à la manœuvre, mais elle perturbe l’attention visuelle dans les autres conditions. Ces résultats permettent d’identifier et de caractériser différents risques inhérents à la RA de surlignage, et de discuter des pistes de conception pour les prendre en compte. / Attention involves distinguishing information which is useful for an activity from that which is not. Augmented reality (AR) highlighting guides this process of information selection by increasing the salience of high-value elements. In such systems, “value” is typically seen as linked to the overall activity of driving (e.g. traffic signs in case of poor visibility, or indications of direction). However, several studies have shown that during an activity eye movements are specific to the immediate goal. AR which does not respect this “natural” prioritizing of information thus risks interfering with information-processing.The first objective of this study is to determine to what extent the allocation of visual attention in automobile driving is focused on information related to a maneuver. The second objective is to study the impact of AR on this allocation of attention.We set up three experiments in which participants viewed static and dynamic driving situations and had to decide whether they could perform a maneuver. We analyzed the variations in allocation of visual attention in relation to the maneuvers in question and the AR conditions using eye movements recording. Our results show that visual attention is strongly linked to cues which permit decision-making, but that it does not overlook cues allowing a general comprehension of the situation. AR optimizes visual attention when it highlights cues related to the maneuver, but it disturbs visual attention in other cases. These findings make it possible to identify and categorize various risks inherent in AR highlighting, and to discuss possible approaches to dealing with them through more effective design.

abstract: Super-Resolution (SR) techniques are widely developed to increase image resolution by fusing several Low-Resolution (LR) images of the same scene to overcome sensor hardware limitations and reduce media impairments in a cost-effective manner. When choosing a solution for the SR problem, there is always a trade-off between computational efficiency and High-Resolution (HR) image quality. Existing SR approaches suffer from extremely high computational requirements due to the high number of unknowns to be estimated in the solution of the SR inverse problem. This thesis proposes efficient iterative SR techniques based on Visual Attention (VA) and perceptual modeling of the human visual system. In the first part of this thesis, an efficient ATtentive-SELective Perceptual-based (AT-SELP) SR framework is presented, where only a subset of perceptually significant active pixels is selected for processing by the SR algorithm based on a local contrast sensitivity threshold model and a proposed low complexity saliency detector. The proposed saliency detector utilizes a probability of detection rule inspired by concepts of luminance masking and visual attention. The second part of this thesis further enhances on the efficiency of selective SR approaches by presenting an ATtentive (AT) SR framework that is completely driven by VA region detectors. Additionally, different VA techniques that combine several low-level features, such as center-surround differences in intensity and orientation, patch luminance and contrast, bandpass outputs of patch luminance and contrast, and difference of Gaussians of luminance intensity are integrated and analyzed to illustrate the effectiveness of the proposed selective SR frameworks. The proposed AT-SELP SR and AT-SR frameworks proved to be flexible by integrating a Maximum A Posteriori (MAP)-based SR algorithm as well as a fast two-stage Fusion-Restoration (FR) SR estimator. By adopting the proposed selective SR frameworks, simulation results show significant reduction on average in computational complexity with comparable visual quality in terms of quantitative metrics such as PSNR, SNR or MAE gains, and subjective assessment. The third part of this thesis proposes a Perceptually Weighted (WP) SR technique that incorporates unequal weighting parameters in the cost function of iterative SR problems. The proposed approach is inspired by the unequal processing of the Human Visual System (HVS) to different local image features in an image. Simulation results show an enhanced reconstruction quality and faster convergence rates when applied to the MAP-based and FR-based SR schemes. / Dissertation/Thesis / Ph.D. Electrical Engineering 2011

Doctor of Philosophy / Physics / Nobel S. Rebello / This dissertation investigates the effects of visual cueing and outcome feedback on students' performance, confidence, and visual attention as they solve conceptual physics problems that contain diagrams. The research investigation had two parts. In the first part of the study, participants solved four sets of conceptual physics problems that contain diagrams; each set contained an initial problem, four isomorphic training problems, a near transfer problem (with a slightly different surface feature as the training problems), and a far transfer problem (with considerably different surface feature as the training problems). Participants in the cued conditions saw visual cues overlaid on the training problem diagrams, while those in the feedback conditions were told if their responses were correct or incorrect. In the second part of the study, the same students solved the near and far transfer problems from the first study two weeks later. We found that the combination of visual cueing and outcome feedback improved performance on the near transfer and delayed near transfer problems compared to the initial problem, with no significant difference between them. Thus, the combination of visual cueing and outcome feedback can promote immediate learning and retention. For students who demonstrated immediate learning and retention on the near and far transfer problems, visual cues improved the automaticity of extracting relevant information from the transfer and delayed transfer problem diagrams, while outcome feedback helped automatize the extraction of problem-relevant information on the delayed far transfer problem diagram only. We also showed that students' reported confidence in solving a problem is positively related to their correctness on the problem, and their visual attention to the relevant information on the problem diagram. The most interesting thing was how changes in confidence occurred due to outcome feedback, which were also related to changes in accuracy and visual attention. The changes in confidence included both reductions in confidence and increases in confidence due to feedback when the student was wrong (first) and right (later). This seems to have led to learning (change in accuracy), and also changes in attentional allocation (more attention to the thematically relevant area).

Existe controvérsia na literatura sobre a possibilidade de divisão espacial da atenção visual. Alguns autores encontraram evidências de que a atenção visual voluntária se divide no espaço. No entanto, a possibilidade de divisão da atenção automática ainda não foi adequadamente investigada. O objetivo deste trabalho foi investigar a possibilidade de uma divisão da atenção visual automática no espaço. Em um primeiro e segundo experimentos testamos a possibilidade da atenção automática se dividir em um mesmo hemicampo (esquerdo ou direito). Em um terceiro e quarto experimentos investigamos a possibilidade de divisão da atenção automática entre os hemicampos visuais (esquerdo e direito). Em conjunto, os resultados deste trabalho sugerem que a atenção automática pode se dividir entre os hemicampos visuais esquerdo e direito, mas em cada um destes hemicampos forma um foco único longo e estreito. / Several studies demonstrated that voluntary visual attention can be divided. The possibility that this also occurs for automatic visual attention was investigated here. In the first and second experiments of this study the possibility of attention division in the same hemifield was examined. In the third and fouth experiments the possibility of attention division between hemifields was examined. The results suggest that automatic visual attention can not divide in the same hemifield but may divide between hemifields.

Há evidências de que a orientação da atenção para o lado direito seria mais eficiente do que para o esquerdo. Outros trabalhos mostram uma assimetria quanto às vias visuais. A via ventral, que realiza processamento de objetos, seria mais ativa no hemisfério esquerdo e a via dorsal, que realiza a localização de estímulos, seria mais ativa no hemisfério direito. A interação entre essas assimetrias foi investigada. Realizamos dois experimentos nos quais a atenção automática dos sujeitos foi mobilizada através do aparecimento de um estímulo visual periférico. O Experimento 1 foi uma tarefa de discriminação de forma, envolvendo a via ventral. O Experimento 2 foi de discriminação de local, envolvendo a via dorsal. Esperávamos a manifestação de uma assimetria comportamental para o lado direito apenas no Experimento 1. No entanto isto ocorreu nos dois experimentos, ou seja, assimetria foi para o lado em que a atenção é preferencialmente mobilizada. O que sugere que o fator determinante para essa manifestação é a assimetria atencional, mas não assimetria das vias visuais. / Recent studies have shown that attention orienting to the right side is more efficient than attention orienting to the left side. Other studies have shown that the two hemispheres process visual stimuli differently: the ventral visual pathway, which processes object shape, is more active in the left hemisphere, while the dorsal visual pathway, which processes object location, is more active in the right hemisphere. This work investigated theses asymmetries. Automatic attention was mobilized by a peripheral visual prime stimulus in two experiments. In Experiment 1 a task requiring shape discrimination was used. It was expected a strong bias to the right side. In Experiment 2 a task requiring location discrimination was used. It was expected now a much smaller bias to the right side. A strong bias to the right side was found in the two experiments. This bias was probably due to the attentional bias to the right side. Interhemispheric sensory asymmetries apparently were not strong enough to influence behavior in the present experimental conditions.

Em tarefas de tempo de reação, uma pista visual pode engatilhar diversos processos, que facitam ou inibem diferentes estágios do processamento do estímulo-alvo. As características da tarefa podem modular os efeitos da pista. Neste estudo, investigamos as possíveis causas dos efeitos inibitórios da pista. Em experimentos em que foi usada uma tarefa de discriminação vai/não-vai, um efeito inibitório foi encontrado quando um valor fixo assincronia entre o início pista e o início do estímulo-alvo (SOA) foi usado, mas não quando o valor de SOA foi variado no decorrer da tarefa. Na tarefa de discriminação de escolha, nenhum efeito initório da pista foi observado, enquanto na tarefa de detecção houve facilitação. Nós propomos que o efeito inibitório na tarefa do tipo vai-não/vai se origina da inibição da tendência de responder automaticamente ao aparecimento da pista. Essa inibição tornaria mais lenta a resposta ao estímulo-alvo subsequente. / In reaction time tasks, a visual cue can trigger several processes and each of these processes can have either a facilitatory or inhibitory influence on different stages of target processing. It has been suggested that task set might modulate cuing effects. In the present study, we investigated the possible causes of inhibitory cueing effects. In experiments involving a go/no-go discrimination task, an inhibitory cueing effect on reaction time was found when a fixed asynchrony between cue onset and target onset (SOA) was used, but not when varied SOAs was used. In a choice discrimination task no inhibitory cueing effect was observed, while in a detection task there was a facilitatory cueing effect. We propose that the inhibitory cueing effect in the go/no-go task arises from the inhibition of an automatic tendency to respond to the cue. This would slow responding to the subsequent target stimulus. Our results suggest that a response inhibition to the cue may contribute to the inhibition of return effect.