User-driven saliency maps for evaluating Region-of-Interest detection

Detection of Region of Interest (ROI) in a video leads to more efficient utilization of bandwidth. This is because any ROIs in a given frame can be encoded in higher quality than the rest of that frame, with little or no degradation of quality from the perception of the viewers. Consequently, it is not necessary to uniformly encode the whole video in high quality. One approach to determine ROIs is to use saliency detectors to locate salient regions. This paper proposes a methodology for obtaining ground truth saliency maps to measure the effectiveness of ROI detection by considering the role of user experience during the labelling process of such maps. User perceptions can be captured and incorporated into the definition of salience in a particular video, taking advantage of human visual recall within a given context. Experiments with two state-of-the-art saliency detectors validate the effectiveness of this approach to validating visual saliency in video. This paper will provide the relevant datasets associated with the experiments.

Online Estimation of Evolving Human Visual Interest

Regions in video streams attracting human interest contribute significantly to human understanding of the video. Being able to predict salient and informative Regions of Interest (ROIs) through a sequence of eye movements is a challenging problem. Applications such as content-aware retargeting of videos to different aspect ratios while preserving informative regions and smart insertion of dialog (closed-caption text) into the video stream can significantly be improved using the predicted ROIs. We propose an interactive human-in-the-loop framework to model eye movements and predict visual saliency into yet-unseen frames. Eye tracking and video content are used to model visual attention in a manner that accounts for important eye-gaze characteristics such as temporal discontinuities due to sudden eye movements, noise, and behavioral artifacts. A novel statistical-and algorithm-based method gaze buffering is proposed for eye-gaze analysis and its fusion with content-based features. Our robust saliency prediction is instantiated for two challenging and exciting applications. The first application alters video aspect ratios on-the-fly using content-aware video retargeting, thus making them suitable for a variety of display sizes. The second application dynamically localizes active speakers and places dialog captions on-the-fly in the video stream. Our method ensures that dialogs are faithful to active speaker locations and do not interfere with salient content in the video stream. Our framework naturally accommodates personalisation of the application to suit biases and preferences of individual users.

Semantic feature-based visual attention model for pedestrian detection

Objective
Pedestrian detection under video surveillance systems has always been a hot topic in computer vision research. These systems are widely used in train stations, airports, large commercial plazas, and other public places. However, pedestrian detection remains difficult because of complex backgrounds. Given its development in recent years, the visual attention mechanism has attracted increasing attention in object detection and tracking research, and previous studies have achieved substantial progress and breakthroughs. We propose a novel pedestrian detection method based on the semantic features under the visual attention mechanism.
Method
The proposed semantic feature-based visual attention model is a spatial-temporal model that consists of two parts: the static visual attention model and the motion visual attention model. The static visual attention model in the spatial domain is constructed by combining bottom-up with top-down attention guidance. Based on the characteristics of pedestrians, the bottom-up visual attention model of Itti is improved by intensifying the orientation vectors of elementary visual features to make the visual saliency map suitable for pedestrian detection. In terms of pedestrian attributes, skin color is selected as a semantic feature for pedestrian detection. The regional and Gaussian models are adopted to construct the skin color model. Skin feature-based visual attention guidance is then proposed to complete the top-down process. The bottom-up and top-down visual attentions are linearly combined using the proper weights obtained from experiments to construct the static visual attention model in the spatial domain. The spatial-temporal visual attention model is then constructed via the motion features in the temporal domain. Based on the static visual attention model in the spatial domain, the frame difference method is combined with optical flowing to detect motion vectors. Filtering is applied to process the field of motion vectors. The saliency of motion vectors can be evaluated via motion entropy to make the selected motion feature more suitable for the spatial-temporal visual attention model.
Result
Standard datasets and practical videos are selected for the experiments. The experiments are performed on a MATLAB R2012a platform. The experimental results show that our spatial-temporal visual attention model demonstrates favorable robustness under various scenes, including indoor train station surveillance videos and outdoor scenes with swaying leaves. Our proposed model outperforms the visual attention model of Itti, the graph-based visual saliency model, the phase spectrum of quaternion Fourier transform model, and the motion channel model of Liu in terms of pedestrian detection. The proposed model achieves a 93% accuracy rate on the test video.
Conclusion
This paper proposes a novel pedestrian method based on the visual attention mechanism. A spatial-temporal visual attention model that uses low-level and semantic features is proposed to calculate the saliency map. Based on this model, the pedestrian targets can be detected through focus of attention shifts. The experimental results verify the effectiveness of the proposed attention model for detecting pedestrians.

Social saliency : visual psychophysics and single-neuron recordings in humans

My thesis studies how people pay attention to other people and the environment. How does the brain figure out what is important and what are the neural mechanisms underlying attention? What is special about salient social cues compared to salient non-social cues? In Chapter I, I review social cues that attract attention, with an emphasis on the neurobiology of these social cues. I also review neurological and psychiatric links: the relationship between saliency, the amygdala and autism. The first empirical chapter then begins by noting that people constantly move in the environment. In Chapter II, I study the spatial cues that attract attention during locomotion using a cued speeded discrimination task. I found that when the motion was expansive, attention was attracted towards the singular point of the optic flow (the focus of expansion, FOE) in a sustained fashion. The more ecologically valid the motion features became (e.g., temporal expansion of each object, spatial depth structure implied by distribution of the size of the objects), the stronger the attentional effects. However, compared to inanimate objects and cues, people preferentially attend to animals and faces, a process in which the amygdala is thought to play an important role. To directly compare social cues and non-social cues in the same experiment and investigate the neural structures processing social cues, in Chapter III, I employ a change detection task and test four rare patients with bilateral amygdala lesions. All four amygdala patients showed a normal pattern of reliably faster and more accurate detection of animate stimuli, suggesting that advantageous processing of social cues can be preserved even without the amygdala, a key structure of the “social brain”. People not only attend to faces, but also pay attention to others’ facial emotions and analyze faces in great detail. Humans have a dedicated system for processing faces and the amygdala has long been associated with a key role in recognizing facial emotions. In Chapter IV, I study the neural mechanisms of emotion perception and find that single neurons in the human amygdala are selective for subjective judgment of others’ emotions. Lastly, people typically pay special attention to faces and people, but people with autism spectrum disorders (ASD) might not. To further study social attention and explore possible deficits of social attention in autism, in Chapter V, I employ a visual search task and show that people with ASD have reduced attention, especially social attention, to target-congruent objects in the search array. This deficit cannot be explained by low-level visual properties of the stimuli and is independent of the amygdala, but it is dependent on task demands. Overall, through visual psychophysics with concurrent eye-tracking, my thesis found and analyzed socially salient cues and compared social vs. non-social cues and healthy vs. clinical populations. Neural mechanisms underlying social saliency were elucidated through electrophysiology and lesion studies. I finally propose further research questions based on the findings in my thesis and introduce my follow-up studies and preliminary results beyond the scope of this thesis in the very last section, Future Directions.

Contextual visual localization: cascaded submap classification, optimized saliency detection, and fast view matching

In this paper, we present a novel coarse-to-fine visual localization approach: contextual visual localization. This approach relies on three elements: (i) a minimal-complexity classifier for performing fast coarse localization (submap classification); (ii) an optimized saliency detector which exploits the visual statistics of the submap; and (iii) a fast view-matching algorithm which filters initial matchings with a structural criterion. The latter algorithm yields fine localization. Our experiments show that these elements have been successfully integrated for solving the global localization problem. Context, that is, the awareness of being in a particular submap, is defined by a supervised classifier tuned for a minimal set of features. Visual context is exploited both for tuning (optimizing) the saliency detection process, and to select potential matching views in the visual database, close enough to the query view.

Neural routing circuits for forming invariant representations of visual objects

This thesis presents a biologically plausible model of an attentional mechanism for forming position- and scale-invariant representations of objects in the visual world. The model relies on a set of control neurons to dynamically modify the synaptic strengths of intra-cortical connections so that information from a windowed region of primary visual cortex (Vl) is selectively routed to higher cortical areas. Local spatial relationships (i.e., topography) within the attentional window are preserved as information is routed through the cortex, thus enabling attended objects to be represented in higher cortical areas within an object-centered reference frame that is position and scale invariant. The representation in V1 is modeled as a multiscale stack of sample nodes with progressively lower resolution at higher eccentricities. Large changes in the size of the attentional window are accomplished by switching between different levels of the multiscale stack, while positional shifts and small changes in scale are accomplished by translating and rescaling the window within a single level of the stack. The control signals for setting the position and size of the attentional window are hypothesized to originate from neurons in the pulvinar and in the deep layers of visual cortex. The dynamics of these control neurons are governed by simple differential equations that can be realized by neurobiologically plausible circuits. In pre-attentive mode, the control neurons receive their input from a low-level "saliency map" representing potentially interesting regions of a scene. During the pattern recognition phase, control neurons are driven by the interaction between top-down (memory) and bottom-up (retinal input) sources. The model respects key neurophysiological, neuroanatomical, and psychophysical data relating to attention, and it makes a variety of experimentally testable predictions.

Perceptual Organization, Figure-Ground, Attention and Saliency

Notions of figure-ground, inside-outside are difficult to define in a computational sense, yet seem intuitively meaningful. We propose that "figure" is an attention-directed region of visual information processing, and has a non-discrete boundary. Associated with "figure" is a coordinate frame and a "frame curve" which helps initiate the shape recognition process by selecting and grouping convex image chunks for later matching- to-model. We show that human perception is biased to see chunks outside the frame as more salient than those inside. Specific tasks, however, can reverse this bias. Near/far, top/bottom and expansion/contraction also behave similarly.

Contextual Influences on Saliency

This article describes a model for including scene/context priors in attention guidance. In the proposed scheme, visual context information can be available early in the visual processing chain, in order to modulate the saliency of image regions and to provide an efficient short cut for object detection and recognition. The scene is represented by means of a low-dimensional global description obtained from low-level features. The global scene features are then used to predict the probability of presence of the target object in the scene, and its location and scale, before exploring the image. Scene information can then be used to modulate the saliency of image regions early during the visual processing in order to provide an efficient short cut for object detection and recognition.

Cortical Dynamics of Contextually-Cued Attentive Visual Learning and Search: Spatial and Object Evidence Accumulation

How do humans use predictive contextual information to facilitate visual search? How are consistently paired scenic objects and positions learned and used to more efficiently guide search in familiar scenes? For example, a certain combination of objects can define a context for a kitchen and trigger a more efficient search for a typical object, such as a sink, in that context. A neural model, ARTSCENE Search, is developed to illustrate the neural mechanisms of such memory-based contextual learning and guidance, and to explain challenging behavioral data on positive/negative, spatial/object, and local/distant global cueing effects during visual search. The model proposes how global scene layout at a first glance rapidly forms a hypothesis about the target location. This hypothesis is then incrementally refined by enhancing target-like objects in space as a scene is scanned with saccadic eye movements. The model clarifies the functional roles of neuroanatomical, neurophysiological, and neuroimaging data in visual search for a desired goal object. In particular, the model simulates the interactive dynamics of spatial and object contextual cueing in the cortical What and Where streams starting from early visual areas through medial temporal lobe to prefrontal cortex. After learning, model dorsolateral prefrontal cortical cells (area 46) prime possible target locations in posterior parietal cortex based on goalmodulated percepts of spatial scene gist represented in parahippocampal cortex, whereas model ventral prefrontal cortical cells (area 47/12) prime possible target object representations in inferior temporal cortex based on the history of viewed objects represented in perirhinal cortex. The model hereby predicts how the cortical What and Where streams cooperate during scene perception, learning, and memory to accumulate evidence over time to drive efficient visual search of familiar scenes.

Artistic rendering of the visual cortex

In this paper we explain the processing in the first layers of the visual cortex by simple, complex and endstopped cells, plus grouping cells for line, edge, keypoint and saliency detection. Three visualisations are presented: (a) an integrated scheme that shows activities of simple, complex and end-stopped cells, (b) artistic combinations of selected activity maps that give an impression of global image structure and/or local detail, and (c) NPR on the basis of a 2D brightness model. The cortical image representations offer many possibilities for non-photorealistic rendering.

Visual attention and perception models for assessing quality in 2D and 3D stereoscopic video

La medida de calidad de vídeo sigue siendo necesaria para definir los criterios que caracterizan una señal que cumpla los requisitos de visionado impuestos por el usuario. Las nuevas tecnologías, como el vídeo 3D estereoscópico o formatos más allá de la alta definición, imponen nuevos criterios que deben ser analizadas para obtener la mayor satisfacción posible del usuario. Entre los problemas detectados durante el desarrollo de esta tesis doctoral se han determinado fenómenos que afectan a distintas fases de la cadena de producción audiovisual y tipo de contenido variado. En primer lugar, el proceso de generación de contenidos debe encontrarse controlado mediante parámetros que eviten que se produzca el disconfort visual y, consecuentemente, fatiga visual, especialmente en lo relativo a contenidos de 3D estereoscópico, tanto de animación como de acción real. Por otro lado, la medida de calidad relativa a la fase de compresión de vídeo emplea métricas que en ocasiones no se encuentran adaptadas a la percepción del usuario. El empleo de modelos psicovisuales y diagramas de atención visual permitirían ponderar las áreas de la imagen de manera que se preste mayor importancia a los píxeles que el usuario enfocará con mayor probabilidad. Estos dos bloques se relacionan a través de la definición del término saliencia. Saliencia es la capacidad del sistema visual para caracterizar una imagen visualizada ponderando las áreas que más atractivas resultan al ojo humano. La saliencia en generación de contenidos estereoscópicos se refiere principalmente a la profundidad simulada mediante la ilusión óptica, medida en términos de distancia del objeto virtual al ojo humano. Sin embargo, en vídeo bidimensional, la saliencia no se basa en la profundidad, sino en otros elementos adicionales, como el movimiento, el nivel de detalle, la posición de los píxeles o la aparición de caras, que serán los factores básicos que compondrán el modelo de atención visual desarrollado. Con el objetivo de detectar las características de una secuencia de vídeo estereoscópico que, con mayor probabilidad, pueden generar disconfort visual, se consultó la extensa literatura relativa a este tema y se realizaron unas pruebas subjetivas preliminares con usuarios. De esta forma, se llegó a la conclusión de que se producía disconfort en los casos en que se producía un cambio abrupto en la distribución de profundidades simuladas de la imagen, aparte de otras degradaciones como la denominada “violación de ventana”. A través de nuevas pruebas subjetivas centradas en analizar estos efectos con diferentes distribuciones de profundidades, se trataron de concretar los parámetros que definían esta imagen. Los resultados de las pruebas demuestran que los cambios abruptos en imágenes se producen en entornos con movimientos y disparidades negativas elevadas que producen interferencias en los procesos de acomodación y vergencia del ojo humano, así como una necesidad en el aumento de los tiempos de enfoque del cristalino. En la mejora de las métricas de calidad a través de modelos que se adaptan al sistema visual humano, se realizaron también pruebas subjetivas que ayudaron a determinar la importancia de cada uno de los factores a la hora de enmascarar una determinada degradación. Los resultados demuestran una ligera mejora en los resultados obtenidos al aplicar máscaras de ponderación y atención visual, los cuales aproximan los parámetros de calidad objetiva a la respuesta del ojo humano. ABSTRACT Video quality assessment is still a necessary tool for defining the criteria to characterize a signal with the viewing requirements imposed by the final user. New technologies, such as 3D stereoscopic video and formats of HD and beyond HD oblige to develop new analysis of video features for obtaining the highest user’s satisfaction. Among the problems detected during the process of this doctoral thesis, it has been determined that some phenomena affect to different phases in the audiovisual production chain, apart from the type of content. On first instance, the generation of contents process should be enough controlled through parameters that avoid the occurrence of visual discomfort in observer’s eye, and consequently, visual fatigue. It is especially necessary controlling sequences of stereoscopic 3D, with both animation and live-action contents. On the other hand, video quality assessment, related to compression processes, should be improved because some objective metrics are adapted to user’s perception. The use of psychovisual models and visual attention diagrams allow the weighting of image regions of interest, giving more importance to the areas which the user will focus most probably. These two work fields are related together through the definition of the term saliency. Saliency is the capacity of human visual system for characterizing an image, highlighting the areas which result more attractive to the human eye. Saliency in generation of 3DTV contents refers mainly to the simulated depth of the optic illusion, i.e. the distance from the virtual object to the human eye. On the other hand, saliency is not based on virtual depth, but on other features, such as motion, level of detail, position of pixels in the frame or face detection, which are the basic features that are part of the developed visual attention model, as demonstrated with tests. Extensive literature involving visual comfort assessment was looked up, and the development of new preliminary subjective assessment with users was performed, in order to detect the features that increase the probability of discomfort to occur. With this methodology, the conclusions drawn confirmed that one common source of visual discomfort was when an abrupt change of disparity happened in video transitions, apart from other degradations, such as window violation. New quality assessment was performed to quantify the distribution of disparities over different sequences. The results confirmed that abrupt changes in negative parallax environment produce accommodation-vergence mismatches derived from the increasing time for human crystalline to focus the virtual objects. On the other side, for developing metrics that adapt to human visual system, additional subjective tests were developed to determine the importance of each factor, which masks a concrete distortion. Results demonstrated slight improvement after applying visual attention to objective metrics. This process of weighing pixels approximates the quality results to human eye’s response.