Les écrans d'ordinateurs sont de très petite taille comparés à celles des jeux de donnés dans de nombreux domaines. Pour pallier au problème de visualisation de grandes quantités de données, les interfaces de navigation multi-échelles rendent possible l'exploration interactive des données, en facilitant la transition entre vues zoomées et dé-zoomées afin de permettre à l'utilisateur d'examiner les informations en détail ou de pouvoir les interpréter dans un contexte plus global. L'étude de ces interfaces est un domaine important de la recherche en Interaction Homme-Machine, et de nombreuses techniques de navigation ont été proposées lors des vingt dernières années. Nous proposons une nouvelle approche de conception pour les interfaces de navigation multi-échelles dîtes content-aware. Cette approche est basée sur l'adaptation dynamique d'éléments de l'interface au contenu de la scène que l'utilisateur est en train de visualiser, permettant ainsi de proposer des représentations plus pertinentes. Nous présentons trois nouvelles techniques de navigation basées sur cette approche de conception, qui montrent comment appliquer celle-ci pour traiter différents problèmes de navigation, aussi bien en 2D qu'en 3D. Nous présentons dans un premier temps Arealens et Pathlens, deux lentilles de grossissement 2D dont la forme va s'adapter à la géométrie des objets d'intérêts afin de proposer une meilleur intégration de la vue zoomée dans son contexte environnant. Une expérience de laboratoire contrôlée de Arealens met en évidence un gain de performance de ce type de lentille par rapport aux lentilles de grossissement classiques pour une tâche de recherche visuelle. Nous introduisons ensuite Gimlens, une lentille de grossissement 3D permettant l'exploration de modèles complexes. Gimlens permet d'identifier rapidement les objets d'intérêt de la scène, d'afficher des vues détaillées de ces objets, de parcourir des orbites autour de ceux-ci et de les mettre en perspective dans leur contexte environnant. L'utilisateur peut également combiner les lentilles pour afficher simultanément différentes vues complémentaires de la scène / Computer screens are very small compared to the size of large information spaces that arise in many domains. The visualization of such datasets requires multiscale navigation capabilities, enabling users to switch between zoomed-in detailed views and zoomed-out contextual views of the data. Designing interfaces that allow users to quickly identify objects of interest, get detailed views of those objects, relate them and put them in a broader spatial context, raise challenging issues. Multi-scale interfaces have been the focus of much research effort over the last twenty years.There are several design approaches to address multiscale navigation issues. In this thesis, we review and categorize these approaches according to their level of content awareness. We identify two main approaches: content-driven, which optimizes interfaces for navigation in specific content; and content-agnostic, that applies to any type of data. We introduce the content-aware design approach, which dynamically adapts the interface to the content. The latter design approach can be used to design multiscale navigation techniques both in 2D or 3D spaces. We introduce Arealens and Pathlens, two content-aware fisheye lenses that dynamically adapt their shape to the underlying content to better preserve the visual aspect of objects of interest. We describe the techniques and their implementation, and report on a controlled experiment that evaluates the usability of Arealens compared to regular fisheye lenses, showing clear performance improvements with the new technique for a multiscale visual search task. We introduce a new distortion-oriented presentation library enabling the design of fisheye lenses featuring several foci of arbitrary shapes. Then, we introduce Gimlens, a multi-view detail-in-context visualization technique that enables users to navigate complex 3D models by drilling holes into their outer layers to reveal objects that are buried into the scene. Gimlens adapts to the geometry of objects of interest so as to better manage visual occlusion problems, selection mechanism and coordination of lenses.

This study explores how alternative scrolling techniques compare to traditional vertical scrolling. It is studied in the context of user experience and information allocation. The alternative scrolling methods are evaluated through a focus group and a controlled experiment. A focus group explores different scrolling techniques and participants' attitudes toward alternative scrolling methods while seeking dissatisfaction with normal scrolling. The controlled experiment delves deeper into the hedonic and pragmatic qualities of scrolling, examining the time it takes for participants to allocate information to be complemented with semi-structured interviews. By delving deeper into the user experience and user preference for scrolling, the study aims to answer the following research question: “How can scrolling techniques alternative to vertical and continuous scrolling benefit users in terms of efficiently allocating specific information and increasing perceived user experience and usability?”. In summary, the study's findings show that the alternative scrolling techniques are not statistically more efficient than normal conventional scrolling. They are, however, viewed as having more hedonic qualities and are generally more preferred from a usability standpoint.

This thesis addresses the challenge of denoising microscopy images captured under low-light conditionswith varying intensity levels. The study compares three deep learning models — N2V, CARE, andRCAN — against the collaborative filter BM4D, which serves as a reference point. The models weretrained on two distinct datasets: Endoplasmic Reticulum and Mitochondria datasets, both acquired witha lattice light-sheet microscope.Results show that BM4D maintains stable performance metrics and delivers superior visual quality,when compared to the noisy input. In contrast, the deep learning models exhibit poor performance onnoisy test images when trained on datasets with non-uniform noise levels. Additionally, a sensitivitycomparison of neural parameter between the same models was made. Revealing that supervised modelsare data-specific to some extent, whereas the self-supervised N2V demonstrates consistent neuralparameters, suggesting lower data specificity. / Denna uppsats tar upp problemet med att reducera brus i mikroskopibilder tagna under svagaljusförhållanden med varierande intensitetsnivåer. Studien jämför tre djupinlärningsmodeller – N2V,CARE och RCAN – mot det kollaborativa filtret BM4D, vilket agerar som en referenspunkt.Modellerna tränades på två olika dataset: Endoplasmic Reticulum och Mitochondria, båda tagna meden selektiv planbelysningsmikroskop (lattice light-sheet microscope).Resultaten visar att BM4D behåller stabila prestationsmått och levererar bättre visuell kvalitet, jämförtmed den brusiga input. Däremot visar djupinlärningsmodellerna bristande prestanda på brusigatestbilder när de tränats på data med icke-enhetliga brusnivåer. Dessutom gjordes enkänslighetsjämförelse av neurala parametrar mellan samma modeller. Detta visade att de övervakademodellerna är specifika för data i viss utsträckning, medan den självövervakade N2V-modellen visarlika neurala parametrar, vilket tyder på lägre dataspecificitet