A multimedia indexing and retrieval framework for multimedia database systems

The main challenges of multimedia data retrieval lie in the effective mapping between low-level features and high-level concepts, and in the individual users' subjective perceptions of multimedia content. ^ The objectives of this dissertation are to develop an integrated multimedia indexing and retrieval framework with the aim to bridge the gap between semantic concepts and low-level features. To achieve this goal, a set of core techniques have been developed, including image segmentation, content-based image retrieval, object tracking, video indexing, and video event detection. These core techniques are integrated in a systematic way to enable the semantic search for images/videos, and can be tailored to solve the problems in other multimedia related domains. In image retrieval, two new methods of bridging the semantic gap are proposed: (1) for general content-based image retrieval, a stochastic mechanism is utilized to enable the long-term learning of high-level concepts from a set of training data, such as user access frequencies and access patterns of images. (2) In addition to whole-image retrieval, a novel multiple instance learning framework is proposed for object-based image retrieval, by which a user is allowed to more effectively search for images that contain multiple objects of interest. An enhanced image segmentation algorithm is developed to extract the object information from images. This segmentation algorithm is further used in video indexing and retrieval, by which a robust video shot/scene segmentation method is developed based on low-level visual feature comparison, object tracking, and audio analysis. Based on shot boundaries, a novel data mining framework is further proposed to detect events in soccer videos, while fully utilizing the multi-modality features and object information obtained through video shot/scene detection. ^ Another contribution of this dissertation is the potential of the above techniques to be tailored and applied to other multimedia applications. This is demonstrated by their utilization in traffic video surveillance applications. The enhanced image segmentation algorithm, coupled with an adaptive background learning algorithm, improves the performance of vehicle identification. A sophisticated object tracking algorithm is proposed to track individual vehicles, while the spatial and temporal relationships of vehicle objects are modeled by an abstract semantic model. ^

Multiple contour extraction from graylevel images using an artificial neural network

For active contour modeling (ACM), we propose a novel self-organizing map (SOM)-based approach, called the batch-SOM (BSOM), that attempts to integrate the advantages of SOM- and snake-based ACMs in order to extract the desired contours from images. We employ feature points, in the form of ail edge-map (as obtained from a standard edge-detection operation), to guide the contour (as in the case of SOM-based ACMs) along with the gradient and intensity variations in a local region to ensure that the contour does not "leak" into the object boundary in case of faulty feature points (weak or broken edges). In contrast with the snake-based ACMs, however, we do not use an explicit energy functional (based on gradient or intensity) for controlling the contour movement. We extend the BSOM to handle extraction of contours of multiple objects, by splitting a single contour into as many subcontours as the objects in the image. The BSOM and its extended version are tested on synthetic binary and gray-level images with both single and multiple objects. We also demonstrate the efficacy of the BSOM on images of objects having both convex and nonconvex boundaries. The results demonstrate the superiority of the BSOM over others. Finally, we analyze the limitations of the BSOM.

Computational modeling and psychophysics in low- and mid-level vision

This thesis addresses a series of topics related to the question of how people find the foreground objects from complex scenes. With both computer vision modeling, as well as psychophysical analyses, we explore the computational principles for low- and mid-level vision.

We first explore the computational methods of generating saliency maps from images and image sequences. We propose an extremely fast algorithm called Image Signature that detects the locations in the image that attract human eye gazes. With a series of experimental validations based on human behavioral data collected from various psychophysical experiments, we conclude that the Image Signature and its spatial-temporal extension, the Phase Discrepancy, are among the most accurate algorithms for saliency detection under various conditions.

In the second part, we bridge the gap between fixation prediction and salient object segmentation with two efforts. First, we propose a new dataset that contains both fixation and object segmentation information. By simultaneously presenting the two types of human data in the same dataset, we are able to analyze their intrinsic connection, as well as understanding the drawbacks of today’s “standard” but inappropriately labeled salient object segmentation dataset. Second, we also propose an algorithm of salient object segmentation. Based on our novel discoveries on the connections of fixation data and salient object segmentation data, our model significantly outperforms all existing models on all 3 datasets with large margins.

In the third part of the thesis, we discuss topics around the human factors of boundary analysis. Closely related to salient object segmentation, boundary analysis focuses on delimiting the local contours of an object. We identify the potential pitfalls of algorithm evaluation for the problem of boundary detection. Our analysis indicates that today’s popular boundary detection datasets contain significant level of noise, which may severely influence the benchmarking results. To give further insights on the labeling process, we propose a model to characterize the principles of the human factors during the labeling process.

The analyses reported in this thesis offer new perspectives to a series of interrelating issues in low- and mid-level vision. It gives warning signs to some of today’s “standard” procedures, while proposing new directions to encourage future research.

Mid-level concept learning with visual contextual ontologies and probabilistic inference for image annotation

To date, automatic recognition of semantic information such as salient objects and mid-level concepts from images is a challenging task. Since real-world objects tend to exist in a context within their environment, the computer vision researchers have increasingly incorporated contextual information for improving object recognition. In this paper, we present a method to build a visual contextual ontology from salient objects descriptions for image annotation. The ontologies include not only partOf/kindOf relations, but also spatial and co-occurrence relations. A two-step image annotation algorithm is also proposed based on ontology relations and probabilistic inference. Different from most of the existing work, we specially exploit how to combine representation of ontology, contextual knowledge and probabilistic inference. The experiments show that image annotation results are improved in the LabelMe dataset.

Images of desire : cognitive models of craving

Cognitive modelling of phenomena in clinical practice allows the operationalisation of otherwise diffuse descriptive terms such as craving or flashbacks. This supports the empirical investigation of the clinical phenomena and the development of targeted treatment interventions. This paper focuses on the cognitive processes underpinning craving, which is recognised as a motivating experience in substance dependence. We use a high-level cognitive architecture, Interacting Cognitive Subsystems (ICS), to compare two theories of craving: Tiffany's theory, centred on the control of automated action schemata, and our own Elaborated Intrusion theory of craving. Data from a questionnaire study of the subjective aspects of everyday desires experienced by a large non-clinical population are presented. Both the data and the high-level modelling support the central claim of the Elaborated Intrusion theory that imagery is a key element of craving, providing the subjective experience and mediating much of the associated disruption of concurrent cognition.

Pattern recognition using invariants defined from higher order spectra : 2-D image inputs

A new algorithm for extracting features from images for object recognition is described. The algorithm uses higher order spectra to provide desirable invariance properties, to provide noise immunity, and to incorporate nonlinearity into the feature extraction procedure thereby allowing the use of simple classifiers. An image can be reduced to a set of 1D functions via the Radon transform, or alternatively, the Fourier transform of each 1D projection can be obtained from a radial slice of the 2D Fourier transform of the image according to the Fourier slice theorem. A triple product of Fourier coefficients, referred to as the deterministic bispectrum, is computed for each 1D function and is integrated along radial lines in bifrequency space. Phases of the integrated bispectra are shown to be translation- and scale-invariant. Rotation invariance is achieved by a regrouping of these invariants at a constant radius followed by a second stage of invariant extraction. Rotation invariance is thus converted to translation invariance in the second step. Results using synthetic and actual images show that isolated, compact clusters are formed in feature space. These clusters are linearly separable, indicating that the nonlinearity required in the mapping from the input space to the classification space is incorporated well into the feature extraction stage. The use of higher order spectra results in good noise immunity, as verified with synthetic and real images. Classification of images using the higher order spectra-based algorithm compares favorably to classification using the method of moment invariants

Modélisation procédurale par composants

Le réalisme des images en infographie exige de créer des objets (ou des scènes) de plus en plus complexes, ce qui entraîne des coûts considérables. La modélisation procédurale peut aider à automatiser le processus de création, à simplifier le processus de modification ou à générer de multiples variantes d'une instance d'objet. Cependant même si plusieurs méthodes procédurales existent, aucune méthode unique permet de créer tous les types d'objets complexes, dont en particulier un édifice complet. Les travaux réalisés dans le cadre de cette thèse proposent deux solutions au problème de la modélisation procédurale: une solution au niveau de la géométrie de base, et l’autre sous forme d'un système général adapté à la modélisation des objets complexes. Premièrement, nous présentons le bloc, une nouvelle primitive de modélisation simple et générale, basée sur une forme cubique généralisée. Les blocs sont disposés et connectés entre eux pour constituer la forme de base des objets, à partir de laquelle est extrait un maillage de contrôle pouvant produire des arêtes lisses et vives. La nature volumétrique des blocs permet une spécification simple de la topologie, ainsi que le support des opérations de CSG entre les blocs. La paramétrisation de la surface, héritée des faces des blocs, fournit un soutien pour les textures et les fonctions de déplacements afin d'appliquer des détails de surface. Une variété d'exemples illustrent la généralité des blocs dans des contextes de modélisation à la fois interactive et procédurale. Deuxièmement, nous présentons un nouveau système de modélisation procédurale qui unifie diverses techniques dans un cadre commun. Notre système repose sur le concept de composants pour définir spatialement et sémantiquement divers éléments. À travers une série de déclarations successives exécutées sur un sous-ensemble de composants obtenus à l'aide de requêtes, nous créons un arbre de composants définissant ultimement un objet dont la géométrie est générée à l'aide des blocs. Nous avons appliqué notre concept de modélisation par composants à la génération d'édifices complets, avec intérieurs et extérieurs cohérents. Ce nouveau système s'avère général et bien adapté pour le partionnement des espaces, l'insertion d'ouvertures (portes et fenêtres), l'intégration d'escaliers, la décoration de façades et de murs, l'agencement de meubles, et diverses autres opérations nécessaires lors de la construction d'un édifice complet.

LGT/VOT tracking performance evaluation of depth images

This paper presents object tracking in depth, RGB and normal-maps images using LGT tracker. The depth and RGB images are rendered using depth imaging plugins. A series of experiments were held to evaluate the tracker performance in tracking objects in different image sequences. The experiments conducted were from the Visual Object Tracking (VOT) challenge that was arranged in association with ICCV'13 The accuracy was chosen as the evaluation measure, where the the tracker's bounding box was compared against the ground truth bounding box. Results show that tracking object using depth images gives better results and is more accurate than tracking using either the RGB or nomal maps images.

Methods for the analysis of multi-scale cell dynamics from fluorescence microscopy images

La embriogénesis es el proceso mediante el cual una célula se convierte en un ser un vivo. A lo largo de diferentes etapas de desarrollo, la población de células va proliferando a la vez que el embrión va tomando forma y se configura. Esto es posible gracias a la acción de varios procesos genéticos, bioquímicos y mecánicos que interaccionan y se regulan entre ellos formando un sistema complejo que se organiza a diferentes escalas espaciales y temporales. Este proceso ocurre de manera robusta y reproducible, pero también con cierta variabilidad que permite la diversidad de individuos de una misma especie. La aparición de la microscopía de fluorescencia, posible gracias a proteínas fluorescentes que pueden ser adheridas a las cadenas de expresión de las células, y los avances en la física óptica de los microscopios han permitido observar este proceso de embriogénesis in-vivo y generar secuencias de imágenes tridimensionales de alta resolución espacio-temporal. Estas imágenes permiten el estudio de los procesos de desarrollo embrionario con técnicas de análisis de imagen y de datos, reconstruyendo dichos procesos para crear la representación de un embrión digital. Una de las más actuales problemáticas en este campo es entender los procesos mecánicos, de manera aislada y en interacción con otros factores como la expresión genética, para que el embrión se desarrolle. Debido a la complejidad de estos procesos, estos problemas se afrontan mediante diferentes técnicas y escalas específicas donde, a través de experimentos, pueden hacerse y confrontarse hipótesis, obteniendo conclusiones sobre el funcionamiento de los mecanismos estudiados. Esta tesis doctoral se ha enfocado sobre esta problemática intentando mejorar las metodologías del estado del arte y con un objetivo específico: estudiar patrones de deformación que emergen del movimiento organizado de las células durante diferentes estados del desarrollo del embrión, de manera global o en tejidos concretos. Estudios se han centrado en la mecánica en relación con procesos de señalización o interacciones a nivel celular o de tejido. En este trabajo, se propone un esquema para generalizar el estudio del movimiento y las interacciones mecánicas que se desprenden del mismo a diferentes escalas espaciales y temporales. Esto permitiría no sólo estudios locales, si no estudios sistemáticos de las escalas de interacción mecánica dentro de un embrión. Por tanto, el esquema propuesto obvia las causas de generación de movimiento (fuerzas) y se centra en la cuantificación de la cinemática (deformación y esfuerzos) a partir de imágenes de forma no invasiva. Hoy en día las dificultades experimentales y metodológicas y la complejidad de los sistemas biológicos impiden una descripción mecánica completa de manera sistemática. Sin embargo, patrones de deformación muestran el resultado de diferentes factores mecánicos en interacción con otros elementos dando lugar a una organización mecánica, necesaria para el desarrollo, que puede ser cuantificado a partir de la metodología propuesta en esta tesis. La metodología asume un medio continuo descrito de forma Lagrangiana (en función de las trayectorias de puntos materiales que se mueven en el sistema en lugar de puntos espaciales) de la dinámica del movimiento, estimado a partir de las imágenes mediante métodos de seguimiento de células o de técnicas de registro de imagen. Gracias a este esquema es posible describir la deformación instantánea y acumulada respecto a un estado inicial para cualquier dominio del embrión. La aplicación de esta metodología a imágenes 3D + t del pez zebra sirvió para desvelar estructuras mecánicas que tienden a estabilizarse a lo largo del tiempo en dicho embrión, y que se organizan a una escala semejante al del mapa de diferenciación celular y con indicios de correlación con patrones de expresión genética. También se aplicó la metodología al estudio del tejido amnioserosa de la Drosophila (mosca de la fruta) durante el cierre dorsal, obteniendo indicios de un acoplamiento entre escalas subcelulares, celulares y supracelulares, que genera patrones complejos en respuesta a la fuerza generada por los esqueletos de acto-myosina. En definitiva, esta tesis doctoral propone una estrategia novedosa de análisis de la dinámica celular multi-escala que permite cuantificar patrones de manera inmediata y que además ofrece una representación que reconstruye la evolución de los procesos como los ven las células, en lugar de como son observados desde el microscopio. Esta metodología por tanto permite nuevas formas de análisis y comparación de embriones y tejidos durante la embriogénesis a partir de imágenes in-vivo. ABSTRACT The embryogenesis is the process from which a single cell turns into a living organism. Through several stages of development, the cell population proliferates at the same time the embryo shapes and the organs develop gaining their functionality. This is possible through genetic, biochemical and mechanical factors that are involved in a complex interaction of processes organized in different levels and in different spatio-temporal scales. The embryogenesis, through this complexity, develops in a robust and reproducible way, but allowing variability that makes possible the diversity of living specimens. The advances in physics of microscopes and the appearance of fluorescent proteins that can be attached to expression chains, reporting about structural and functional elements of the cell, have enabled for the in-vivo observation of embryogenesis. The imaging process results in sequences of high spatio-temporal resolution 3D+time data of the embryogenesis as a digital representation of the embryos that can be further analyzed, provided new image processing and data analysis techniques are developed. One of the most relevant and challenging lines of research in the field is the quantification of the mechanical factors and processes involved in the shaping process of the embryo and their interactions with other embryogenesis factors such as genetics. Due to the complexity of the processes, studies have focused on specific problems and scales controlled in the experiments, posing and testing hypothesis to gain new biological insight. However, methodologies are often difficult to be exported to study other biological phenomena or specimens. This PhD Thesis is framed within this paradigm of research and tries to propose a systematic methodology to quantify the emergent deformation patterns from the motion estimated in in-vivo images of embryogenesis. Thanks to this strategy it would be possible to quantify not only local mechanisms, but to discover and characterize the scales of mechanical organization within the embryo. The framework focuses on the quantification of the motion kinematics (deformation and strains), neglecting the causes of the motion (forces), from images in a non-invasive way. Experimental and methodological challenges hamper the quantification of exerted forces and the mechanical properties of tissues. However, a descriptive framework of deformation patterns provides valuable insight about the organization and scales of the mechanical interactions, along the embryo development. Such a characterization would help to improve mechanical models and progressively understand the complexity of embryogenesis. This framework relies on a Lagrangian representation of the cell dynamics system based on the trajectories of points moving along the deformation. This approach of analysis enables the reconstruction of the mechanical patterning as experienced by the cells and tissues. Thus, we can build temporal profiles of deformation along stages of development, comprising both the instantaneous events and the cumulative deformation history. The application of this framework to 3D + time data of zebrafish embryogenesis allowed us to discover mechanical profiles that stabilized through time forming structures that organize in a scale comparable to the map of cell differentiation (fate map), and also suggesting correlation with genetic patterns. The framework was also applied to the analysis of the amnioserosa tissue in the drosophila’s dorsal closure, revealing that the oscillatory contraction triggered by the acto-myosin network organized complexly coupling different scales: local force generation foci, cellular morphology control mechanisms and tissue geometrical constraints. In summary, this PhD Thesis proposes a theoretical framework for the analysis of multi-scale cell dynamics that enables to quantify automatically mechanical patterns and also offers a new representation of the embryo dynamics as experienced by cells instead of how the microscope captures instantaneously the processes. Therefore, this framework enables for new strategies of quantitative analysis and comparison between embryos and tissues during embryogenesis from in-vivo images.

Towards automatic tree crown detection and delineation in spectral feature space using PCNN and morphological reconstruction

The application of object-based approaches to the problem of extracting vegetation information from images requires accurate delineation of individual tree crowns. This paper presents an automated method for individual tree crown detection and delineation by applying a simplified PCNN model in spectral feature space followed by post-processing using morphological reconstruction. The algorithm was tested on high resolution multi-spectral aerial images and the results are compared with two existing image segmentation algorithms. The results demonstrate that our algorithm outperforms the other two solutions with the average accuracy of 81.8%.

Learning articulated motion structures with Bayesian Networks

This paper presents a general methodology for learning articulated motions that, despite having non-linear correlations, are cyclical and have a defined pattern of behavior Using conventional algorithms to extract features from images, a Bayesian classifier is applied to cluster and classify features of the moving object. Clusters are then associated in different frames and structure learning algorithms for Bayesian networks are used to recover the structure of the motion. This framework is applied to the human gait analysis and tracking but applications include any coordinated movement such as multi-robots behavior analysis.

Toward a more robust facial expression recognition in occluded images using randomly sampled Gabor based templates

Occlusion is a big challenge for facial expression recognition (FER) in real-world situations. Previous FER efforts to address occlusion suffer from loss of appearance features and are largely limited to a few occlusion types and single testing strategy. This paper presents a robust approach for FER in occluded images and addresses these issues. A set of Gabor based templates is extracted from images in the gallery using a Monte Carlo algorithm. These templates are converted into distance features using template matching. The resulting feature vectors are robust to occlusion. Occluded eyes and mouth regions and randomly places occlusion patches are used for testing. Two testing strategies analyze the effects of these occlusions on the overall recognition performance as well as each facial expression. Experimental results on the Cohn-Kanade database confirm the high robustness of our approach and provide useful insights about the effects of occlusion on FER. Performance is also compared with previous approaches.

In silico experimental modeling of cancer treatment

In silico experimental modeling of cancer involves combining findings from biological literature with computer-based models of biological systems in order to conduct investigations of hypotheses entirely in the computer laboratory. In this paper, we discuss the use of in silico modeling as a precursor to traditional clinical and laboratory research, allowing researchers to refine their experimental programs with an aim to reducing costs and increasing research efficiency. We explain the methodology of in silico experimental trials before providing an example of in silico modeling from the biomathematical literature with a view to promoting more widespread use and understanding of this research strategy.

The influence of colour and sound on neuronal activation during visual object naming

This paper investigates how neuronal activation for naming photographs of objects is influenced by the addition of appropriate colour or sound. Behaviourally, both colour and sound are known to facilitate object recognition from visual form. However, previous functional imaging studies have shown inconsistent effects. For example, the addition of appropriate colour has been shown to reduce antero-medial temporal activation whereas the addition of sound has been shown to increase posterior superior temporal activation. Here we compared the effect of adding colour or sound cues in the same experiment. We found that the addition of either the appropriate colour or sound increased activation for naming photographs of objects in bilateral occipital regions and the right anterior fusiform. Moreover, the addition of colour reduced left antero-medial temporal activation but this effect was not observed for the addition of object sound. We propose that activation in bilateral occipital and right fusiform areas precedes the integration of visual form with either its colour or associated sound. In contrast, left antero-medial temporal activation is reduced because object recognition is facilitated after colour and form have been integrated.

A Distributed Robotic Vision Service

Robotic vision is limited by line of sight and onboard camera capabilities. Robots can acquire video or images from remote cameras, but processing additional data has a computational burden. This paper applies the Distributed Robotic Vision Service, DRVS, to robot path planning using data outside line-of-sight of the robot. DRVS implements a distributed visual object detection service to distributes the computation to remote camera nodes with processing capabilities. Robots request task-specific object detection from DRVS by specifying a geographic region of interest and object type. The remote camera nodes perform the visual processing and send the high-level object information to the robot. Additionally, DRVS relieves robots of sensor discovery by dynamically distributing object detection requests to remote camera nodes. Tested over two different indoor path planning tasks DRVS showed dramatic reduction in mobile robot compute load and wireless network utilization.