京都大学 / 新制・課程博士 / 博士(情報学) / 甲第24741号 / 情博第829号 / 新制||情||139(附属図書館) / 京都大学大学院情報学研究科数理工学専攻 / (主査)教授 山下 信雄, 教授 太田 快人, 教授 永持 仁 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

In the form of images and videos, visual content has always attracted considerable interest and attention to itself since the early days of the computer era. Although, due to the high density of information in such contents, it has always been challenging to generate, process and broadcast videos and images. These challenges grew along with the demand for higher quality content and attained the research community's attention to themselves. Even though many works have been done by researchers and engineers in academic and industrial environments, the demand for high-quality content introduces new constraints on the quality, performance (speed) and energy consumption. This thesis focuses on a couple of image and video processing applications and introduces new approaches and tweaks to improve the performance and save resources while keeping the functionality intact. In the first part, we target Interferometric Synthetic Aperture Radar (InSAR), an imaging technique used by satellites to capture the earth's surface. Many algorithms have been developed to extract useful information, such as height and displacement, from such images. However, the sheer size of these images, along with the complexity of most of these algorithms, lead to very long processing time and resource utilization. In this work, we take one of the dominant algorithms used for almost every In-SAR application, Phase Unwrapping, and introduce an approach to gain up to 6.5 times speedups. We evaluated our method on InSAR images taken by the Radarsat-2 sensor and showed its impact on a real-world application. In the second part of this thesis, we look at a prevalent application, video streaming. These days video streaming dominates the internet traffic, so any slight improvement in terms of energy consumption or resource utilization will make a sizable difference. Although the streamers use various encoding techniques, the quality of experience of the clients prevents them from overplaying these techniques. On the other hand, there has been a growing interest in another venture of research which focuses on developing techniques that aim to restore the quality of the videos that have been subjected to compression. Although these techniques are used by many users on the receiver side, the streamers often ignore their capabilities. In our work, we introduce an approach that makes the streamer aware of the capabilities of the receiver and utilizes that awareness to reduce the cost of transmission without compromising the end user's quality of experience. We demonstrated the technique and proved our concept by applying it to the HEVC encoding standard and JCT-VC dataset. / Graduate

No description available.

Worldwide, around 17% of food waste produced in high income countries is constituted of dairy products. In Sweden, Dairy Company A has been dealing with food waste production throughout all their sites. Although it already practices actions such regarding food waste minimization such as biogas generation and animal food, reduction and prevention practices are still an area of opportunity. The aim of this study is to explore alternatives for food waste prevention in Company A CSE dairy supply chain and generate recommendations that help the company achieve a more sustainable supply chain.  This study furthermore proposes a 5-step research methodology to identify food waste causes and drivers, including data gathering mainly through interviews and analysis based on proposed frameworks in relevant literature. It was found that activities related to systematic root cause identification and information storage and sharing play a key role in improving existing practices that aim to reduce and minimize food waste production from a supply planning perspective.  The research identified and recommended not only to raise awareness, which is key to reducing food waste, but to put into a broader context how decisions taken by any planner could be influencing the sustainability of the supply chain. This could be done through the adoption of process-driven mechanisms to share relevant information as well as food waste root cause identification. / Över hela världen utgörs cirka 17 % av matavfallet som produceras i höginkomstländer av mejeriprodukter. I Sverige har Mejeriföretag A sysslat med produktion av matavfall på alla sina anläggningar. Även om det redan tillämpar åtgärder som när det gäller minimering av matsvinn, såsom generering av biogas och djurfoder, är metoder för reduktion och förebyggande fortfarande ett område med möjligheter. Syftet med denna studie är att utforska alternativ för att förebygga matsvinn i Company A CSE-mejeriförsörjningskedjan och generera rekommendationer som hjälper företaget att uppnå en mer hållbar försörjningskedja.  Denna studie föreslår dessutom en 5-stegs forskningsmetodik för att identifiera orsaker och drivkrafter för matsvinn, inklusive datainsamling främst genom intervjuer och analyser baserade på föreslagna ramverk i relevant litteratur. Det visade sig att aktiviteter relaterade till systematisk rotorsaksidentifiering och informationslagring och -delning spelar en nyckelroll för att förbättra befintliga metoder som syftar till att minska och minimera produktionen av matavfall ur ett utbudsplaneringsperspektiv.  Forskningen identifierade och rekommenderade inte bara att öka medvetenheten, vilket är nyckeln till att minska matsvinnet, utan för att sätta in i ett bredare sammanhang hur beslut som fattas av alla planerare kan påverka leveranskedjans hållbarhet. Detta skulle kunna göras genom antagandet av processdrivna mekanismer för att dela relevant information samt identifiering av rotorsak för matavfall.

With the increasing pervasiveness of digital cameras, the Internet, and social networking, there is a growing need to catalog and analyze large collections of photos and videos. In this dissertation, we explore unconstrained still-image and video-based face recognition in real-world scenarios, e.g. social photo sharing and movie trailers, where people of interest are recognized and all others are ignored. In such a scenario, we must obtain high precision in recognizing the known identities, while accurately rejecting those of no interest. Recent advancements in face recognition research has seen Sparse Representation-based Classification (SRC) advance to the forefront of competing methods. However, its drawbacks, slow speed and sensitivity to variations in pose, illumination, and occlusion, have hindered its wide-spread applicability. The contributions of this dissertation are three-fold: 1. For still-image data, we propose a novel Linearly Approximated Sparse Representation-based Classification (LASRC) algorithm that uses linear regression to perform sample selection for l1-minimization, thus harnessing the speed of least-squares and the robustness of SRC. On our large dataset collected from Facebook, LASRC performs equally to standard SRC with a speedup of 100-250x. 2. For video, applying the popular l1-minimization for face recognition on a frame-by-frame basis is prohibitively expensive computationally, so we propose a new algorithm Mean Sequence SRC (MSSRC) that performs video face recognition using a joint optimization leveraging all of the available video data and employing the knowledge that the face track frames belong to the same individual. Employing MSSRC results in a speedup of 5x on average over SRC on a frame-by-frame basis. 3. Finally, we make the observation that MSSRC sometimes assigns inconsistent identities to the same individual in a scene that could be corrected based on their visual similarity. Therefore, we construct a probabilistic affinity graph combining appearance and co-occurrence similarities to model the relationship between face tracks in a video. Using this relationship graph, we employ random walk analysis to propagate strong class predictions among similar face tracks, while dampening weak predictions. Our method results in a performance gain of 15.8% in average precision over using MSSRC alone.

The thesis discusses a novel off-line and on-line learning approach for Fully Recurrent Neural Networks (FRNNs). The most popular algorithm for training FRNNs, the Real Time Recurrent Learning (RTRL) algorithm, employs the gradient descent technique for finding the optimum weight vectors in the recurrent neural network. Within the framework of the research presented, a new off-line and on-line variation of RTRL is presented, that is based on the Gauss-Newton method. The method itself is an approximate Newton's method tailored to the specific optimization problem, (non-linear least squares), which aims to speed up the process of FRNN training. The new approach stands as a robust and effective compromise between the original gradient-based RTRL (low computational complexity, slow convergence) and Newton-based variants of RTRL (high computational complexity, fast convergence). By gathering information over time in order to form Gauss-Newton search vectors, the new learning algorithm, GN-RTRL, is capable of converging faster to a better quality solution than the original algorithm. Experimental results reflect these qualities of GN-RTRL, as well as the fact that GN-RTRL may have in practice lower computational cost in comparison, again, to the original RTRL.

Artificial Intelligence has recently attracted attention due to its rapid advancement in various industries such as the healthcare and finance industry. The intersection between AI and entrepreneurship is still being studied, and this study explores the impact of AI-tools on startups, with a focus on Swedish startups. The study explores the utilization of AI- tools to optimize their operations or capture new opportunities. It also examines their alignment to The Lean Startup Development, designed to help entrepreneurs to navigate through challenges they face when launching a product or a service. The primary contribution is of qualitative nature, using semi-structured interviews with individuals from startups implementing AI-technologies. Interpretation of the data is done through thematic analysis, which involves identifying themes and core categories.  The startups use the AI-tools for strategic internal planning and operations. The findings suggest that the AI- tools are commonly used to minimize costs, automate certain tasks, saving time to focus more on complex tasks and thereby enhancing efficiency which gradually leads to strengthened competitiveness. Interestingly, the participating startups show a consideration for ethical risks making more careful decisions on the information provided by the AI-tools.

No description available.

No description available.

No description available.