The advancement of computer architecture and chip design has set the stage for the deep learning revolution by supplying enormous computational power. In general, deep learning is built upon neural networks that can be regarded as a universal approximator of any mathematical function. In contrast to model-based machine learning where the representative features are designed by human engineers, deep learning enables the automatic discovery of desirable feature representations based on a data-driven manner. In this thesis, the applications of deep learning to visual content processing and analysis are discussed. For visual content processing, two novel approaches, named LCVSR and RawVSR, are proposed to address the common issues in the filed of Video Super-Resolution (VSR). In LCVSR, a new mechanism based on local dynamic filters via Locally Connected (LC) layers is proposed to implicitly estimate and compensate motions. It avoids the errors caused by the inaccurate explicit estimation of flow maps. Moreover, a global refinement network is proposed to exploit non-local correlations and enhance the spatial consistency of super-resolved frames. In RawVSR, the superiority of camera raw data (where the primitive radiance information is recorded) is harnessed to benefit the reconstruction of High-Resolution (HR) frames. The developed network is in line with the real imaging pipeline, where the super-resolution process serves as a pre-processing unit of ISP. Moreover, a Successive Deep Inference (SDI) module is designed in accordance with the architectural principle suggested by a canonical decomposition result for Hidden Markov Model (HMM) inference, and a reconstruction module is built with elaborately designed Attention based Residual Dense Blocks (ARDBs). For visual content analysis, a new approach, named PSCC-Net, is proposed to detect and localize image manipulations. It consists of two paths: a top-down path that extracts the local and global features from an input image, and a bottom-up path that first distinguishes manipulated images from pristine ones via a detection head, and then localizes forged regions via a progressive mechanism, where manipulation masks are estimated from small scales to large ones, each serving as a prior of the next-scale estimation. Moreover, a Spatio-Channel Correlation Module (SCCM) is proposed to capture both spatial and channel-wise correlations among extracted features, enabling the network to cope with a wide range of manipulation attacks. Extensive experiments validate that the proposed methods in this thesis have achieved the SOTA results and partially addressed the existing issues in previous works. / Dissertation / Doctor of Philosophy (PhD)

As the major component of the plant cell wall, cellulose is produced by all plant species at an annual rate of over a hundred billion tonnes, making it the most abundant biopolymer on Earth. The hierarchical assembly of cellulose glucan chains into crystalline fibrils, bundles and higher-order networks endows cellulose with its high mechanical strength, but makes it challenging to breakdown and produce cellulose-based nanomaterials and renewable biofuels. In order to fully leverage the potential of cellulose as a sustainable resource, it is important to study the supramolecular structure and hydrolysis of this biomaterial from the nano- to the microscale. In this thesis, we develop new chemical strategies for fluorescently labeling cellulose and employ advanced imaging techniques to study its supramolecular structure at the singlefibril level. The developed labeling method provides a simple and efficient route for fluorescently tagging cellulose nanomaterials with commercially available dyes, yielding high degrees of labeling without altering the native properties of the nanocelluloses. This allowed the preparation of samples that were optimal for super-resolution fluorescence microscopy (SRFM), which was used to provide for the first time, a direct visualization of periodic disorder along the crystalline structure of individual cellulose fibrils. The alternating disordered and crystalline structure observed in SFRM was corroborated with time-lapsed acid hydrolysis experiments to propose a mechanism for the acid hydrolysis of cellulose fibrils. To gain insight on the ultrastructural origin of these regions, we applied a correlative super-resolution light and electron microscopy (SR-CLEM) workflow and observed that the disordered regions were associated nanostructural defects present along cellulose fibrils. Overall, the findings presented in this work provide significant advancements in our understanding of the hierarchical structure and depolymerization of cellulose, which will be useful for the development of new and efficient ways of breaking down this polymer for the production of renewable nanomaterials and bio-based products like biofuels and bioplastics. / Thesis / Doctor of Philosophy (PhD) / In this dissertation, we have studied in unprecedented detail the structure of cellulose – a polymer that is found in every plant. As the main structural component of the plant cell wall, cellulose endows trees with their strength and resilience while storing sunlight energy in its chemical bonds. Since plant biomass represents eighty percent of all living matter on Earth, cellulose is an abundant resource that can be used to produce sustainable and environmentally benign nanomaterials and bioproducts, like biofuels and bioplastics. Our ability to use cellulose as a renewable source of structural materials and energy is intimately linked to our capacity to break apart its tight structural packing. Deconstructing cellulose into various forms demands that we understand the multi-level organization of its structure and the susceptible regions within it. To gain this information, in this thesis we develop new labeling methods and apply state-of-the-art microscopy tools to directly visualize the arrangement of cellulose fibrils at the nanoscale (comparable to 1/10,000 the width of a human hair) and study their breakdown by acid treatment. The findings presented in this work furthers our fundamental understanding of the natural structure of cellulose, which has important implications on the development of industrial strategies to break down this abundant and renewable biomaterial.

Chez les bactéries, les pôles cellulaires servent de domaines subcellulaires impliqués dans plusieurs processus cellulaires. Chez l’agent pathogène du choléra, Vibrio cholerae, en forme de bâtonnet incurvé, le pole contenant l’unique flagelle est impliqué dans la virulence. La protéine d’ancrage polaire HubP interagit avec plusieurs ATPases telles que ParA1 (ségrégation des chromosomes), ParC (localisation polaire du système de chimiotaxie) et FlhG (biosynthèse des flagelles), organisant ainsi l'identité polaire de V. cholerae. Cependant, les mécanismes moléculaires exacts de cet ancrage polaire doivent encore être élucidés. L’objectif de cette thèse est d’établir une vue d'ensemble de l'organisation de pôle cellulaire ce qui implique le mécanisme d’orchestration des différentes fonctions cellulaires par l’identification de l’ensemble des partenaires d'interaction de HubP ainsi que la cartographie fine du pôle cellulaire par microscopie à super résolution (PALM). Afin d’identifier de nouveaux partenaires d'interaction de HubP, j'ai étudié la différence de composition en protéines polaires entre les contextes HubP+ et HubP-. La composition en protéines polaires a été quantifiée de manière relative et absolue en ajoutant des Tag isobares aux protéines extraites de mini-cellules. Ces mini-cellules correspondent des petits compartiments cellulaires issus d’un évènement de division anormal proche du pole et sont enrichies en protéines polaires. Parmi ~800 protéines identifiées, ~ 80 protéines ont été considérées comme enrichies en contexte HubP+ incluant de nombreuses protéines attendues (FlhG, ParC et en aval des protéines de chimiotaxie). J'ai étudié la localisation de 14 protéines par microscopie à fluorescence et pu révéler 4 nouvelles protéines présentant une localisation polaire dépendant de HubP : VbrX, VbrY, et 2 protéines hypothétiques MotV et MotW. La délétion de motV et motW provoque un défaut significatif de propagation dans une gélose molle suggérant une implication dans la chimiotaxie et/ou la motilité. Alors que la microscopie électronique a montré que les deux mutants ont bien un flagelle polaire unique, le suivi-vidéo de leur déplacement a révélé que les deux mutants présentaient des défauts de nage assez distincts: ∆motV est plutôt affecté dans le changement de direction et ∆motW dans la vitesse de déplacement. Des expériences de microscopie fluorescente ont montré que MotV, MotW et HubP présentaient des dynamiques de localisation polaire distinctes au cours du cycle cellulaire. Pour une observation fine du pôle cellulaire par PALM, de nouveaux outils d’analyse d’image à haut débit étaient exigés. La précision des contours des petites cellules bactériennes faiblement contrastées n’est pas suffisante par l’observation en fond clair, j'ai développé une nouvelle technique de marquage avec des protéines fluorescentes photo-activables pour un tracé précis de la membrane interne ou du périplasme. En outre, nous avons créé un logiciel utilisant Matlab appelé Vibio qui intègre le contour de cellule et la liste des molécules obtenues par microscopie à super résolution. La capacité d’analyse à haut débit du logiciel permet d’étudier la distribution des molécules de l’échelle de la cellule unique à une population en orientant les cellules par leur courbure longitudinale. J’ai pu révéler que HubP est principalement localisé du côté convexe du pôle de la cellule, tandis que ses partenaires se situaient principalement au milieu du pôle. Mon travail de thèse a révélé avec succès de nouveaux partenaires d'interaction de HubP et la fonction de certaines protéines dans la motilité cellulaire. J'ai développé une nouvelle technique de microscopie pour une localisation subpolaire précise qui fonctionne bien pour l'analyse d'images PALM dans Vibio. J’ai ainsi pu faire progresser les connaissances de l’orchestration des fonctions polaires chez V. cholerae. / In rod shaped bacteria, cell poles serve as important subcellular domains involved in several cellular processes including motility, chemotaxis, protein secretion, antibiotic resistance, and chromosome segregation. In the cholera pathogen Vibrio cholerae, vibrioid rod shape and single polarized flagellum involve in the virulence. Polar landmark protein HubP was shown to interact with multiple ATPases, such as ParA1 (chromosome segregation), ParC (polar localization of chemotaxis apparatus), and FlhG (flagella biosynthesis), thus organizing the polar identity of V. cholerae by tethering proteins to cell pole. However, the exact molecular mechanisms are yet to be elucidated. In this thesis, I tackled to unveil comprehensive view of the cell pole organization which implies the orchestration of different cellular functions, by identifying further interaction partners of HubP as well as drawing conceivable picture of the cell pole by super-resolution photoactivated localization microscopy. To identify new interaction partners of HubP, I used minicells in which cell poles were enriched as they derived from cell division near the cell pole. Difference in protein composition between HubP+ and HubP- minicells were examined by isobaric tags for relative and absolute quantitation. Among ~800 proteins identified, ~80 proteins were considered to be enriched in HubP+ minicells including many expected proteins (FlhG, ParC and downstream chemotaxis proteins). I chose 14 proteins to investigate their subcellular localization with fluorescent microscopy. In conclusion, I discovered 4 proteins that showed polar localization in a HubP-dependent manner. These proteins are VbrX, VbrY, and 2 hypothetical proteins MotV and MotW. ∆motV and ∆motW showed significant defect in a diameter of travel in soft agar plate that suggesting the possible involvement in chemotaxis and/or motility. Whereas electron microscopy showed that both mutants possess intact monotrichous flagellum, video-tracking revealed that the two mutants showed rather distinct defects during swimming: MotV is rather turning mutant while MotW is a speed mutant. Fluorescent microscopy experiments indicated that MotV, MotW and HubP showed distinct polar dynamics over cell cycle. For fine-scale observation of the cell pole by PALM, it was appreciated that novel tools for high-throughput analysis was demanded. Since brightfield images are not sufficient to have accurate contours of small and low contrast bacterial cells, I developed new labeling technique with photoactivatable fluorescent proteins for precise outlining at either inner membrane or periplasm. Furthermore, we created Matlab-based software called Vibio which integrates cell outline and the list of molecules obtained by super-resolution microscopy. High-throughput capability of the software enabled to analyze distribution of detected molecules from single cell to whole bunch of cells in a manner that cells are oriented by cell curvature. These allowed me to discover that HubP is mostly lopsided at the convex side of the cell pole, while its partners mostly located middle of the pole. Altogether, I successfully unveiled 4 novel interaction partners of HubP. I revealed of the function of hypothetical proteins that are involved in cell motility. I developed new labeling technique for precise polar localization that works well for PALM image analysis in Vibio. Therefore, I observed precise polar localization of HubP and other polar proteins.

<p>Optical imaging methods are limited by the wavelength of light that they use and the amount of scatter that must be imaged through. Super-resolution imaging and sensing methods are those that bypass or mitigate such restrictions. Two super-resolution approaches are presented here using spatially or temporally structured light. Temporal intermittence or blinking of fluorescent emitters is exploited for localization through significant depths of heavy scatter to high resolution, and an efficient algorithm for doing so is presented.</p> <p>Such temporal structure of emission allows far greater resolution than previous comparable imaging methods, providing opportunities in biophotonics and environmental sensing. Spatial structure can be imposed on coherent light that passes through a heavily scattering medium, in the form of a speckle pattern. Speckle intensity correlations are sensitive to the motion of a moving object obscured by scatter, and we demonstrate that this scatter can act as an analyzer, enhancing this sensitivity as the amount of scatter increases. This increased sensitivity is studied using random matrix theory, and eigenchannel analysis is proposed as an explanation. Simulations demonstrate that a randomly scattering analyzer can give sub-wavelength geometric information about a translated, hidden object. Relative motion of structured illumination is explored, with simulations and mathematical analysis demonstrating far-subwavelength sensitivity using moving fields with multiple different types of structure. This work could enable a new approach for material inspection and characterization, and provide improvements in microscopy. </p>

No description available.

No description available.

Fluorescence microscopy is an important tool for biological research. However, the resolution of a standard fluorescence microscope is limited by diffraction, which makes it difficult to observe small details of a specimen’s structure. We have developed two fluorescence microscopy methods that achieve resolution beyond the classical diffraction limit. The first method represents an extension of localisation microscopy. We used nonnegative matrix factorisation (NMF) to model a noisy dataset of highly overlapping fluorophores with intermittent intensities. We can recover images of individual sources from the optimised model, despite their high mutual overlap in the original dataset. This allows us to consider blinking quantum dots as bright and stable fluorophores for localisation microscopy. Moreover, NMF allows recovery of sources each having a unique shape. Such a situation can arise, for example, when the sources are located in different focal planes, and NMF can potentially be used for three dimensional superresolution imaging. We discuss the practical aspects of applying NMF to real datasets, and show super-resolution images of biological samples labelled with quantum dots. It should be noted that this technique can be performed on any wide-field epifluorescence microscope equipped with a camera, which makes this super-resolution method very accessible to a wide scientific community. The second optical microscopy method we discuss in this thesis is a member of the growing family of structured illumination techniques. Our main goal is to apply structured illumination to thick fluorescent samples generating a large out-of-focus background. The out-of-focus fluorescence background degrades the illumination pattern, and the reconstructed images suffer from the influence of noise. We present a combination of structured illumination microscopy and line scanning. This technique reduces the out-of-focus fluorescence background, which improves the quality of the illumination pattern and therefore facilitates reconstruction. We present super-resolution, optically sectioned images of a thick fluorescent sample, revealing details of the specimen’s inner structure. In addition, in this thesis we also discuss a theoretical resolution limit for noisy and pixelated data. We correct a previously published expression for the so-called fundamental resolution measure (FREM) and derive FREM for two fluorophores with intermittent intensity. We show that the intensity intermittency of the sources (observed for quantum dots, for example) can increase the “resolution” defined in terms of FREM.

No description available.

<p>Optical microscopy plays a crucial role in the biological sciences for its ability to enable visualization of biological samples at sub-cellular levels. Many imaging subdivisions exist under this umbrella of general microscopy, and each are tailored towards specific design, contrast, and visualization constraints. Standard examples that have found widespread use include dark-field, phase-contrast, holographic, and fluorescent microscopies. However, a critical factor that physically limits the optical resolution of general microscopy is diffraction. Unfortunately, this “diffraction-limit” can prevent visualization of significant biologically relevant structures, which in turn can limit biological insights. In response to such a limit, several works have advanced the field of sub-diffraction resolution imaging, which consist of optical imaging techniques that seek to achieve imaging resolutions beyond that which is allowed by the diffraction-limit. This set of techniques can largely be divided into two classes. The first class of sub-diffraction techniques is targeted towards cases where the sample is coherently illuminated and diffracts into the imaging system’s aperture. For such cases, synthetic aperture (SA) is a popular choice and operates by using oblique illuminations to spatiotemporally synthesize a wider frequency support into the image than allowed by the diffraction limit. The second class of sub-diffraction techniques, often referred to as "super-resolution" techniques, typically utilize specialized fluorophores with either photoswitching or depletion capabilities. Photoactivated localization microscopy (PALM) is a super-resolution example that localizes photoswitchable fluorophores to sub-diffraction resolutions per acquisition, before combining into a final super-resolved image. Stimulated emission depletion (STED) is another super-resolution example that spatially modulates its excitation to narrow its optical point-spread-function. Unfortunately, SA and fluorescent super-resolution techniques are generally incompatible for sub-diffraction resolution fluorescent and coherent imaging, respectively – thus, a multimodal sub-diffraction imaging solution compatible with both coherent and fluorescent imaging has remained elusive. </p><p> In this dissertation, we demonstrate that structured illumination (SI) is a sub-diffraction technique compatible with both diffractive and fluorescent imaging. We first develop the theoretical framework that extends SI to coherent imaging and experimentally demonstrate SI’s capabilities for 2D sub-diffraction resolution imaging of coherently diffractive samples. Sub-diffraction resolution imaging based on scattering intensity and transmission-based quantitative-phase (QP) are shown. In addition, we show extend SI to 3D coherent imaging, and show applications of this towards 3D QP and refractive-index (RI) tomography. Finally, we show multimodal applications of SI that allow sub-diffraction resolution fluorescent and coherent imaging, which has great potential utility for the biological sciences.</p> / Dissertation

Alors que la microscopie super-résolue a apporté une amélioration considérable en imagerie des assemblages moléculaires dans les milieux biologiques à l'échelle nanométrique, son extension à l'imagerie de l'orientation moléculaire, utilisant l'anisotropie de fluorescence, n'a pas encore été complètement explorée. Apporter une information sur l'orientation moléculaire à l'échelle nanométrique aurait un intérêt considérable pour la compréhension des fonctions biologiques. Dans cette thèse, nous proposons une technique de microscopie super-résolution résolue en polarisation, capable d'imager les comportements d'orientation moléculaire dans des environnements statiques et dynamiques, dans le but de rapporter une information structurale à l'échelle de la molécule unique et à des échelles spatiales nanométriques. En utilisant la microscopie par reconstruction stochastique (dSTORM) en combinaison avec une détection polarisée, des images d'anisotropie de fluorescence sont reconstruites avec une résolution spatiale de quelques dizaines de nanomètres. Nous analysons numériquement le principe de la méthode en combinaison avec des modèles des mécanismes d'orientation moléculaire. Enfin, nous proposons une technique alternative basée sur l'émission de molécules uniques en fluctuations stochastiques: l'imagerie super-résolue polarisée par fluctuations (polar-SOFI), et comparons cette approche avec la précédente. Nous illustrons les deux techniques pour l'imagerie de l'ordre moléculaire dans des fibres de stress d'actine et de tubuline dans des cellules fixées, des fibres d'ADN et des fibrilles d'amyloïde à base d'insuline. / While super-resolution microscopy has brought a significant improvement in nanoscale imaging of molecular assemblies in biological media, its extension to imaging molecular orientation using fluorescence anisotropy has not yet been fully explored. Providing orientational order information at the nanoscale would be of considerable interest for the understanding of biological functions since they are intrinsically related to structural fundamental processes such as in protein clustering in cell membranes, supra-molecular polymerization or aggregation. In this thesis, we propose a super-resolution polarization-resolved microscopy technique able to image molecular orientation behaviors in static and dynamic environments, in order to report structural information at the single molecule level and at nanometric spatial scale. Using direct Stochastic Optical Reconstruction Microscopy (dSTORM) in combination with polarized detection, fluorescence anisotropy images are reconstructed at a spatial resolution of a few tens of nanometers. We analyze numerically the principle of the method in combination with models for orientational order mechanisms, and provide conditions for which this information can be retrieved with high precision in biological samples based on fibrillar structures. Finally, we propose an alternative technique based on stochastic fluctuations of single molecules: polarized super-resolution optical fluctuation imaging (polar-SOFI), and compare this approach with the previous one. We illustrate both techniques on molecular order imaging in actin stress fibers and tubulin fibers in fixed cells, DNA fibers and insulin amyloid fibrils.