In this thesis we give a method of synthesising novel views using only existing views and a set of corresponding points across those views. The method described is based on the assumption that the images have been obtained under affine imaging conditions. This has the advantage that the mathematics needed is simpler than in the perspective case but still allows us to synthesise realistic views. We explore the multiview relationships that exist between corresponding points in three affine images. It is shown that, by careful consideration of where the measurement errors occur, it is possible to improve the accuracy of these multi-view relationships by solving a total least squares problem. We demonstrate how the affine multi-view relationships can be used to encode and accurately reconstruct existing views as a linear combination of two (or more) basis views. In order to synthesise completely novel views it is necessary to parameterise either the multi-view relationships or the positions of the control points in each of the views. We describe a method that can be used to parameterise either the control points or the co-efficients of the affine multi-view relationships. The parameterisation method uses a set of sample views to parameterise the control points (or the multiview relationships) in terms of two parameters. These two parameters can then be varied and used to determine the positions of the control points in novel views. Once the positions of the control points in the novel view are known it is possible to render the image using a combination of the intensities in each of the basis views. The method used to parameterise the existing views allows us to both interpolate between the views and also to extrapolate away from the views. We would expect the quality of the synthesised view to deteriorate as the novel view is extrapolated away from the original set of images. In this thesis we discuss the possibility of using the structure within the total least squares solution combined with the parameterisation described above to determine a limit of extrapolation.

In last two decades, semiconductor nanostructures, such as quantum wells, wires and dots, have been recognised as sources and detectors of radiation in the mid- and far-infrared region of the spectrum. Much of a success has been obtained with quantum well based intraband devices, such as quantum cascade lasers and quantum well infrared photodetectors. However due to longer carrier lifetimes in quantum dots, it is expected that optoelectronic devices based on intraband transitions in self-assembled quantum dots would have superior performance to their quantum well counterparts. In order to fully exploit this prospect, appropriate theoretical models describing electronic, optical and transport properties of the active region of these devices need to be developed, which was the subject of this thesis. It was shown how symmetry of the dot shape can be exploited to efficiently calculate the energy levels within the framework of the multiband envelope function method. The implementation of the method in the plane wave representation of the Hamiltonian eigenvalue problem and the results of its application to square based pyramidal InAs/GaAs quantum dots and hexagonal III-nitride quantum dots were given. A semiclassical model of intraband carrier dynamics in quantum dots was then developed and applied to design an optically pumped long wavelength mid-infrared laser based on intersublevel transitions in InAs/GaAs quantum dots. Two orders of magnitude lower pumping flux was predicted than in similar quantum well based devices. Next, simulations of the optical absorption spectrum in the existing quantum dot infrared photodetector structures were performed. A special emphasis was put into quantum dots-in-a-well structures and explanation of the effect of well width on the detection wavelength. A theory of transport in quantum dot infrared photodetectors starting from the energy levels and wavefunctions obtained by solving the envelope Hamiltonian, yielding as output the device characteristics such as dark current and responsivity, was then developed. The comparison with experimental data available in the literature was made, yielding a good agreement. Finally, the theory of electron transport through arrays of closely stacked quantum dots, where coherent and polaronic effects become important, therefore requiring the treatment within the formalism of the nonequilibrium Green's functions, rather than the semiclassical approach, was presented. A design of a structure promising to act as a terahertz quantum dot cascade laser was given.

No description available.

VIDEO segmentation facilitates e±cient video indexing and navigation in large digital video archives. It is an important process in a content-based video indexing and retrieval (CBVIR) system. Many automated solutions performed seg- mentation by utilizing information about the \facts" of the video. These \facts" come in the form of labels that describe the objects which are captured by the cam- era. This type of solutions was able to achieve good and consistent results for some video genres such as news programs and informational presentations. The content format of this type of videos is generally quite standard, and automated solutions were designed to follow these format rules. For example in [1], the presence of news anchor persons was used as a cue to determine the start and end of a meaningful news segment. The same cannot be said for video genres such as movies and feature films. This is because makers of this type of videos utilized different filming techniques to design their videos in order to elicit certain affective response from their targeted audience. Humans usually perform manual video segmentation by trying to relate changes in time and locale to discontinuities in meaning [2]. As a result, viewers usually have doubts about the boundary locations of a meaningful video segment due to their different affective responses. This thesis presents an entirely new view to the problem of high level video segmentation. We developed a novel probabilistic method for affective level video content analysis and segmentation. Our method had two stages. In the first stage, a®ective content labels were assigned to video shots by means of a dynamic bayesian 0. Abstract 3 network (DBN). A novel hierarchical-coupled dynamic bayesian network (HCDBN) topology was proposed for this stage. The topology was based on the pleasure- arousal-dominance (P-A-D) model of a®ect representation [3]. In principle, this model can represent a large number of emotions. In the second stage, the visual, audio and a®ective information of the video was used to compute a statistical feature vector to represent the content of each shot. Affective level video segmentation was achieved by applying spectral clustering to the feature vectors. We evaluated the first stage of our proposal by comparing its emotion detec- tion ability with all the existing works which are related to the field of a®ective video content analysis. To evaluate the second stage, we used the time adaptive clustering (TAC) algorithm as our performance benchmark. The TAC algorithm was the best high level video segmentation method [2]. However, it is a very computationally intensive algorithm. To accelerate its computation speed, we developed a modified TAC (modTAC) algorithm which was designed to be mapped easily onto a field programmable gate array (FPGA) device. Both the TAC and modTAC algorithms were used as performance benchmarks for our proposed method. Since affective video content is a perceptual concept, the segmentation per- formance and human agreement rates were used as our evaluation criteria. To obtain our ground truth data and viewer agreement rates, a pilot panel study which was based on the work of Gross et al. [4] was conducted. Experiment results will show the feasibility of our proposed method. For the first stage of our proposal, our experiment results will show that an average improvement of as high as 38% was achieved over previous works. As for the second stage, an improvement of as high as 37% was achieved over the TAC algorithm.

No description available.

A Mobile Laser Scanning System (MLSS) is a kinematic platform combining different sensors, namely: GPS, IMU and laser scanner. These sensors are integrated and synchronised to a common time base providing 3D geo-referenced data. MLSS is used in several areas; such as 3D urban and landscape modelling for visualisation in planning and road design, simulations for environmental management, and to support land use decision-making. The accuracy of 3D geo-referenced points, achieved via Mobile Laser Scanning (MLS) under normal conditions, can reach the level of 3cm. However, this accuracy tends to be degraded in urban areas, because of trajectory errors of the laser scanner (IMU drift due to the limited availability of GPS signal). This, also, can be attributed to the difficulty of matching natural features in the point cloud. Previous researches have tried to overcome the problems in urban laser scanning by focusing on enhancing the performance of the navigation system (NGS). This can be costly and may not achieve the high accuracy level required for some engineering application. When the navigation solution is degraded, the accuracy of the point cloud results will be degraded. Using different data sources is another way to improve accuracy in urban areas. For example using airborne LiDAR, terrestrial imagery, and unmanned aerial vehicle (UAV) but these are very time consuming as well as costly compared to MLS systems. Targets are used in a number of ways in MLS and are often chosen from natural detail points. These can be difficult to define, particularly when high accuracy requirements need to be met, for example, when matching scans together or fitting scans to existing surveys as used in this project, and calibrating the system. The accuracy of MLS in the urban area was tested using three methods, namely ground control points (GCPs), surface to surface comparing, and additional source of data. Also, the effect of range, incidence angle (IA), resolution and brightness on different types of targets (sphere, cone, pyramid and flat target) was studied to explore the optimal target design. Moreover, an algorithm for automatic target detection was developed to detect the optimal target. Then, for each target in the point cloud, the centre/apex was calculated using least squares surface fitting. Tests show that the accuracy of 3D coordinates, obtained from MLS in an urban area is about 2-5 cm. Tests also show that using targets with MLS can improve the quality of results reaching 5 mm levels of accuracy even in the urban area, based on the use of check points to assess the quality and reliability of the outputs Almost all work on this project was carried out using the software packages available at the Nottingham Geospatial Institute (NGI) and MLS data provided by 3D Laser Mapping Ltd. (3DLM). Two terrestrial laser scanners, namely: HDS 3000 and Faro Focus3D have been used for testing the designed targets. The findings of this research will contribute easy, cost effective and improved accuracies in MLS data. This enhances usefulness in applications, such as change detection, deformation monitoring, cultural heritage and the process of 3D modelling, particularly in urban areas.

Recent years have seen an explosion of user-generated, untagged multimedia data, generating a need for efficient search and retrieval of this data. The predominant method for content-based tagging is through manual annotation. Consequently, automatic tagging is currently the subject of intensive research. However, it is clear that the process will not be fully automated in the foreseeable future. We propose to involve the user and investigate methods for implicit tagging, wherein users' responses to the multimedia content are analysed in order to generate descriptive tags. We approach this problem through the modalities of facial expressions and EEG signals. We investigate tag validation and affective tagging using EEG signals. The former relies on the detection of event-related potentials triggered in response to the presentation of invalid tags alongside multimedia material. We demonstrate significant differences in users' EEG responses for valid versus invalid tags, and present results towards single-trial classification. For affective tagging, we propose methodologies to map EEG signals onto the valence-arousal space and perform both binary classification as well as regression into this space. We apply these methods in a real-time affective recommendation system. We also investigate the analysis of facial expressions for implicit tagging. This relies on a dynamic texture representation using non-rigid registration that we first evaluate on the problem of facial action unit recognition. We present results on well-known datasets (with both posed and spontaneous expressions) comparable to the state of the art in the field. Finally, we present a multi-modal approach that fuses both modalities for affective tagging. We perform classification in the valence-arousal space based on these modalities and present results for both feature-level and decision-level fusion. We demonstrate improvement in the results when using both modalities, suggesting the modalities contain complementary information.

Witnessing the omnipresence of digital video media, the research community has raised the question of its meaningful use and management. Stored in immense multimedia databases, digital videos need to be retrieved and structured in an intelligent way, relying on the content and the rich semantics involved. Current Content Based Video Indexing and Retrieval systems face the problem of the semantic gap between the simplicity of the available visual features and the richness of user semantics. This work focuses on the issues of efficiency and scalability in video indexing and retrieval to facilitate a video representation model capable of semantic annotation. A highly efficient algorithm for temporal analysis and key-frame extraction is developed. It is based on the prediction information extracted directly from the compressed domain features and the robust scalable analysis in the temporal domain. Furthermore, a hierarchical quantisation of the colour features in the descriptor space is presented. Derived from the extracted set of low-level features, a video representation model that enables semantic annotation and contextual genre classification is designed. Results demonstrate the efficiency and robustness of the temporal analysis algorithm that runs in real time maintaining the high precision and recall of the detection task. Adaptive key-frame extraction and summarisation achieve a good overview of the visual content, while the colour quantisation algorithm efficiently creates hierarchical set of descriptors. Finally, the video representation model, supported by the genre classification algorithm, achieves excellent results in an automatic annotation system by linking the video clips with a limited lexicon of related keywords.

Le rayonnement émis par un humain peut être détecté par un capteur infrarouge travaillant dans la bande 8-14µm. Les caméras basées sur ce type de détecteur, initialement réservées aux applications militaires, sont en développement constant grâce à la demande croissante d’applications commerciales telles que la détection de personnes ou la thermographie. Les détecteurs « non refroidis » constitués de matrices microbolomètriques représentent aujourd’hui près de 95% du marché industriel mais demeurent néanmoins trop couteux pour des applications grand-public. Pour remédier à ce problème, le travail reporté dans ce mémoire est relatif à la conception et la réalisation d’un système de vision infrarouge intégrant un capteur thermique matriciel développé au laboratoire et principalement destiné à la détection de personnes. Contrairement aux matrices bolométriques basées sur une variation de résistance électrique, ce capteur fonctionnant par effet thermoélectrique ne nécessite pas d’alimentation et l’interface électronique associée est très simple. Une attention particulière a été portée sur la diminution des coûts de fabrication par réduction des dimensions des pixels thermoélectriques associés à un objectif équipé d’une lentille de Fresnel en polyéthylène afin de permettre l’implantation de ce type de système dans l’habitat individuel. Des séries de microcapteurs 5x5, 7x7, 10x10 pixels ont été développées durant cette thèse ainsi qu’un imageur complet utilisant une matrice de 7x7 pixels. Le modèle mathématique mis au point durant ce travail montre que les dimensions des capteurs peuvent être encore réduites pour obtenir un nombre de pixels plus conséquent sans augmentation du coût. / The thermal radiation emitted by human body can be detected in the 8-14 µm range by an infra-red sensor. Cameras imagined from this type of detectors were initially reserved for army personal usage but the market is in a large growth due to commercial applications like thermography or human detection. Thermal detectors microbolometers are the most spread technologies with 95% of the market but some others are showing interesting results as thermoelectrical arrays. This document explains the realisation of an infrared camera that can be used for human detection using a technology developed in the group. Contrary to the bolometric matrices based on a variation of electrical resistance, this sensor working by thermoelectric effect does not require power supply and the associated electronic interface is very simple. A special attention was related to the reduction in the manufacturing costs by decreasing the size of the thermoelectric pixels and using a polyethylene Fresnel lens in order to use this type of system in the individual housing. Series of 5x5, 7x7, 10x10 pixels microsensors were modelled and realized during this thesis as well as a complete imaging device using of 7x7 pixels array. The mathematical model developed during this work shows that a further increase in length reduction is possible if more pixels are needed.

La maîtrise de l’augmentation des doses de rayonnements ionisants en imagerie médicale constitue un objectif prioritaire des exigences réglementaires européennes et nationales. Cette radioprotection passe notamment par le besoin constant de développement de dispositifs de dosimétrie appropriés, permettant l’évaluation de la dose délivrée. Aujourd’hui, les dispositifs de dosimétrie utilisés en radiologie, en radiothérapie ou en imagerie nucléaire ne sont pas adaptés aux contraintes de la radiologie interventionnelle. Dans le cadre de cette thèse, nous proposons d’étudier si l’usage de nouveaux matériaux et des nanotechnologies, non conventionnels au domaine médical, peut contribuer au développement de solutions de mesure de dose radiotransparentes, avec pour cas d’étude la radiologie interventionnelle. Notre approche a consisté en le développement d’un nouveau modèle permettant de prédire le comportement électromagnétique et le caractère radiotransparent d’un matériau. La radiotransparence et la visibilité sur les images radiographiques qui sont d’habitude traitées au cas par cas, sont abordées d’un point de vue plus général en reproduisant toute la chaîne de production de l’image. Ce modèle a mis en perspective l’intérêt du PEDOT:PSS dans la dosimétrie des rayons X. Parallèlement, nous avons considéré les MOSFET comme indicateurs de mesure de dose en radiologie interventionnelle, utilisés généralement en radiothérapie où les énergies mises en jeu sont bien plus élevées. L’originalité de cet axe est d’étudier des MOSFET issus de l’électronique, non dédiés à la dosimétrie. / Controlling the radiation dose in medical imaging is seen by the authorities as a national and European priority. This is more specifically achieved through the development of appropriate dosimetry devices, ensuring accurate patient dose monitoring. Today dosimetry devices used in radiology, radiotherapy or nuclear imaging are not adapted to the constraints of interventional radiology. In this thesis, we propose to study if new materials and nanotechnology can contribute to the development of radiotransparent dose measurement solution, especially in interventional radiology. Our approach included developing a model that can help predict the electromagnetic behavior and radiolucent features of materials. Radiolucency and visibility on radiographic images which are usually addressed on a case-by-case basis, are approached from a more general perspective by reproducing the all image production. The model emphasizes the interest of PEDOT:PSS in the x-ray dosimetry. At the same time, we considered MOSFET as indicators of dose measurement in interventional radiology. These transistors are usually employed in radiotherapy, in which involved energies are much higher. The originality of this part is to study MOSFET used in electronics and not dedicated to dosimetry.