This tutorial covers biomedical image reconstruction, from the foundational concepts of system modeling and direct reconstruction to modern sparsity and learning-based approaches. Imaging is a critical tool in biological research and medicine, and most imaging systems necessarily use an image-reconstruction algorithm to create an image; the design of these algorithms has been a topic of research since at least the 1960's. In the last few years, machine learning-based approaches have shown impressive performance on image reconstruction problems, triggering a wave of enthusiasm and creativity around the paradigm of learning. Our goal is to unify this body of research, identifying common principles and reusable building blocks across decades and among diverse imaging modalities. We first describe system modeling, emphasizing how a few building blocks can be used to describe a broad range of imaging modalities. We then discuss reconstruction algorithms, grouping them into three broad generations. The first are the classical direct methods, including Tikhonov regularization; the second are the variational methods based on sparsity and the theory of compressive sensing; and the third are the learning-based (also called data-driven) methods, especially those using deep convolutional neural networks. There are strong links between these generations: classical (first-generation) methods appear as modules inside the latter two, and the former two are used to inspire new designs for learning-based (third-generation) methods. As a result, a solid understanding of all of three generations is necessary for the design of state-of-the-art algorithms.
{"title":"Biomedical Image Reconstruction: From the Foundations to Deep Neural Networks","authors":"Michael T. McCann, M. Unser","doi":"10.1561/2000000101","DOIUrl":"https://doi.org/10.1561/2000000101","url":null,"abstract":"This tutorial covers biomedical image reconstruction, from the foundational concepts of system modeling and direct reconstruction to modern sparsity and learning-based approaches. \u0000Imaging is a critical tool in biological research and medicine, and most imaging systems necessarily use an image-reconstruction algorithm to create an image; the design of these algorithms has been a topic of research since at least the 1960's. In the last few years, machine learning-based approaches have shown impressive performance on image reconstruction problems, triggering a wave of enthusiasm and creativity around the paradigm of learning. Our goal is to unify this body of research, identifying common principles and reusable building blocks across decades and among diverse imaging modalities. \u0000We first describe system modeling, emphasizing how a few building blocks can be used to describe a broad range of imaging modalities. We then discuss reconstruction algorithms, grouping them into three broad generations. The first are the classical direct methods, including Tikhonov regularization; the second are the variational methods based on sparsity and the theory of compressive sensing; and the third are the learning-based (also called data-driven) methods, especially those using deep convolutional neural networks. There are strong links between these generations: classical (first-generation) methods appear as modules inside the latter two, and the former two are used to inspire new designs for learning-based (third-generation) methods. As a result, a solid understanding of all of three generations is necessary for the design of state-of-the-art algorithms.","PeriodicalId":12340,"journal":{"name":"Found. Trends Signal Process.","volume":"2013 1","pages":"283-359"},"PeriodicalIF":0.0,"publicationDate":"2019-01-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89516482","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Lianlin Li, M. Hurtado, F. Xu, Bing Zhang, T. Jin, Tie Jun Xui, M. Stevanovic, A. Nehorai
The low-dimensional-model-based electromagnetic imaging is an emerging member of the big family of computational imaging, by which the low-dimensional models of underlying signals are incorporated into both data acquisition systems and reconstruction algorithms for electromagnetic imaging, in order to improve the imaging performance and break the bottleneck of existing electromagnetic imaging methodologies. Over the past decade, we have witnessed profound impacts of the low-dimensional models on electromagnetic imaging. However, the low-dimensional-model-based electromagnetic imaging remains at its early stage, and many Lianlin Li, Martin Hurtado, Feng Xu, Bing Chen Zhang, Tian Jin, Tie Jun Cui, Marija Nikolic Stevanovic and Arye Nehorai (2018), “A Survey on the LowDimensional-Model-based Electromagnetic Imaging”, : Vol. 12, No. 2, pp 107–199. DOI: 10.1561/2000000103.
{"title":"A Survey on the Low-Dimensional-Model-based Electromagnetic Imaging","authors":"Lianlin Li, M. Hurtado, F. Xu, Bing Zhang, T. Jin, Tie Jun Xui, M. Stevanovic, A. Nehorai","doi":"10.1561/2000000103","DOIUrl":"https://doi.org/10.1561/2000000103","url":null,"abstract":"The low-dimensional-model-based electromagnetic imaging is an emerging member of the big family of computational imaging, by which the low-dimensional models of underlying signals are incorporated into both data acquisition systems and reconstruction algorithms for electromagnetic imaging, in order to improve the imaging performance and break the bottleneck of existing electromagnetic imaging methodologies. Over the past decade, we have witnessed profound impacts of the low-dimensional models on electromagnetic imaging. However, the low-dimensional-model-based electromagnetic imaging remains at its early stage, and many Lianlin Li, Martin Hurtado, Feng Xu, Bing Chen Zhang, Tian Jin, Tie Jun Cui, Marija Nikolic Stevanovic and Arye Nehorai (2018), “A Survey on the LowDimensional-Model-based Electromagnetic Imaging”, : Vol. 12, No. 2, pp 107–199. DOI: 10.1561/2000000103.","PeriodicalId":12340,"journal":{"name":"Found. Trends Signal Process.","volume":"102 1","pages":"107-199"},"PeriodicalIF":0.0,"publicationDate":"2018-06-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75980788","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This review addresses the role of synchronization in the radio localization problem, and provides a comprehensive overview of recent developments suitable for current and future practical implementations. The material is intended for both, theoreticians and practitioners, and is written to be accessible to novices, while covering state-of-the-art topics, of interest to advanced researchers of localization and synchronization systems. Several widely-used radio localization systems, such as GPS and cellular localization, rely on time-of-flight measurements of data-bearing signals to determine inter-radio distances. For such measurements to be meaningful, accurate synchronization is required. While existing systems use a highly synchronous infrastructure, such as GPS where satellites are equipped with atomic clocks or cellular localization where base stations are GPS synchronized, most other wireless networks do not have an sufficiently accurate common notion of time across the nodes. Synchronization, either at link or network level, thus has a principal role in localization systems. This role is expected to become more important in view of recent trends in high-precision and distributed localization, as well as future communication standards, such as 5G indoor localization when access points can not be externally synchronized. Since synchronization is generally treated separately from localization, there is a need to harmonize these two fundamental problems, especially in the decentralized network context. In this monograph, we revisit the role of synchronization in radio localization and provide an exposition of its relation to the general network localization problem. After an introduction of basic concepts, models, and network inference methods, we contrast two-step approaches with single-step (simultaneous) synchronization and localization. These approaches are discussed in terms of their methodology and fundamental limitations. Our focus is on techniques that consider practical relevant clock, delay, and measurement models in order to guide the reader from physical observations to statistical estimation techniques. The presented methods apply to networks with asynchronous localization infrastructure and/or to cooperative ad-hoc networks.
{"title":"Synchronization and Localization in Wireless Networks","authors":"B. Etzlinger, H. Wymeersch","doi":"10.1561/2000000096","DOIUrl":"https://doi.org/10.1561/2000000096","url":null,"abstract":"This review addresses the role of synchronization in the radio localization problem, and provides a comprehensive overview of recent developments suitable for current and future practical implementations. The material is intended for both, theoreticians and practitioners, and is written to be accessible to novices, while covering state-of-the-art topics, of interest to advanced researchers of localization and synchronization systems. Several widely-used radio localization systems, such as GPS and cellular localization, rely on time-of-flight measurements of data-bearing signals to determine inter-radio distances. For such measurements to be meaningful, accurate synchronization is required. While existing systems use a highly synchronous infrastructure, such as GPS where satellites are equipped with atomic clocks or cellular localization where base stations are GPS synchronized, most other wireless networks do not have an sufficiently accurate common notion of time across the nodes. Synchronization, either at link or network level, thus has a principal role in localization systems. This role is expected to become more important in view of recent trends in high-precision and distributed localization, as well as future communication standards, such as 5G indoor localization when access points can not be externally synchronized. Since synchronization is generally treated separately from localization, there is a need to harmonize these two fundamental problems, especially in the decentralized network context. In this monograph, we revisit the role of synchronization in radio localization and provide an exposition of its relation to the general network localization problem. After an introduction of basic concepts, models, and network inference methods, we contrast two-step approaches with single-step (simultaneous) synchronization and localization. These approaches are discussed in terms of their methodology and fundamental limitations. Our focus is on techniques that consider practical relevant clock, delay, and measurement models in order to guide the reader from physical observations to statistical estimation techniques. The presented methods apply to networks with asynchronous localization infrastructure and/or to cooperative ad-hoc networks.","PeriodicalId":12340,"journal":{"name":"Found. Trends Signal Process.","volume":"110 1","pages":"1-106"},"PeriodicalIF":0.0,"publicationDate":"2018-03-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81737221","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Massive multiple-input multiple-output MIMO is one of themost promising technologies for the next generation of wirelesscommunication networks because it has the potential to providegame-changing improvements in spectral efficiency SE and energyefficiency EE. This monograph summarizes many years ofresearch insights in a clear and self-contained way and providesthe reader with the necessary knowledge and mathematical toolsto carry out independent research in this area. Starting froma rigorous definition of Massive MIMO, the monograph coversthe important aspects of channel estimation, SE, EE, hardwareefficiency HE, and various practical deployment considerations.From the beginning, a very general, yet tractable, canonical systemmodel with spatial channel correlation is introduced. This modelis used to realistically assess the SE and EE, and is later extendedto also include the impact of hardware impairments. Owing tothis rigorous modeling approach, a lot of classic "wisdom" aboutMassive MIMO, based on too simplistic system models, is shownto be questionable.
{"title":"Massive MIMO Networks: Spectral, Energy, and Hardware Efficiency","authors":"Emil Björnson, J. Hoydis, L. Sanguinetti","doi":"10.1561/2000000093","DOIUrl":"https://doi.org/10.1561/2000000093","url":null,"abstract":"Massive multiple-input multiple-output MIMO is one of themost promising technologies for the next generation of wirelesscommunication networks because it has the potential to providegame-changing improvements in spectral efficiency SE and energyefficiency EE. This monograph summarizes many years ofresearch insights in a clear and self-contained way and providesthe reader with the necessary knowledge and mathematical toolsto carry out independent research in this area. Starting froma rigorous definition of Massive MIMO, the monograph coversthe important aspects of channel estimation, SE, EE, hardwareefficiency HE, and various practical deployment considerations.From the beginning, a very general, yet tractable, canonical systemmodel with spatial channel correlation is introduced. This modelis used to realistically assess the SE and EE, and is later extendedto also include the impact of hardware impairments. Owing tothis rigorous modeling approach, a lot of classic \"wisdom\" aboutMassive MIMO, based on too simplistic system models, is shownto be questionable.","PeriodicalId":12340,"journal":{"name":"Found. Trends Signal Process.","volume":"259 1","pages":"154-655"},"PeriodicalIF":0.0,"publicationDate":"2018-01-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"77140742","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Video Coding is the second part of the two-part monograph Fundamentals of Source and Video Coding by Wiegand and Schwarz. This part describes the application of the techniques described in the first part to video coding. In doing so it provides a description of the fundamentals concepts of video coding and, in particular, the signal processing in video encoders and decoders.
{"title":"Video Coding: Part II of Fundamentals of Source and Video Coding","authors":"T. Wiegand, H. Schwarz","doi":"10.1561/2000000078","DOIUrl":"https://doi.org/10.1561/2000000078","url":null,"abstract":"Video Coding is the second part of the two-part monograph Fundamentals of Source and Video Coding by Wiegand and Schwarz. This part describes the application of the techniques described in the first part to video coding. In doing so it provides a description of the fundamentals concepts of video coding and, in particular, the signal processing in video encoders and decoders.","PeriodicalId":12340,"journal":{"name":"Found. Trends Signal Process.","volume":"29 1","pages":"1-346"},"PeriodicalIF":0.0,"publicationDate":"2016-12-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73651878","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Despite the different nature of financial engineering and electrical engineering, both areas are intimately connected on a mathematical level. The foundations of financial engineering lie on the statistical analysis of numerical time series and the modeling of the behavior of the financial markets in order to perform predictions and systematically optimize investment strategies. Similarly, the foundations of electrical engineering, for instance, wireless communication systems, lie on statistical signal processing and the modeling of communication channels in order to perform predictions and systematically optimize transmission strategies. Both foundations are the same in disguise. It is often the case in science that the same or very similar methodologies are developed and applied independently in different areas. A Signal Processing Perspective of Financial Engineering is about investment in financial assets treated as a signal processing and optimization problem. It explores such connections and capitalizes on the existing mathematical tools developed in wireless communications and signal processing to solve real-life problems arising in the financial markets in an unprecedented way. A Signal Processing Perspective of Financial Engineering provides straightforward and systematic access to financial engineering for researchers in signal processing and communications so that they can understand problems in financial engineering more easily and may even apply signal processing techniques to handle some financial problems.
{"title":"A Signal Processing Perspective of Financial Engineering","authors":"Yiyong Feng, D. Palomar","doi":"10.1561/2000000072","DOIUrl":"https://doi.org/10.1561/2000000072","url":null,"abstract":"Despite the different nature of financial engineering and electrical engineering, both areas are intimately connected on a mathematical level. The foundations of financial engineering lie on the statistical analysis of numerical time series and the modeling of the behavior of the financial markets in order to perform predictions and systematically optimize investment strategies. Similarly, the foundations of electrical engineering, for instance, wireless communication systems, lie on statistical signal processing and the modeling of communication channels in order to perform predictions and systematically optimize transmission strategies. Both foundations are the same in disguise. It is often the case in science that the same or very similar methodologies are developed and applied independently in different areas. A Signal Processing Perspective of Financial Engineering is about investment in financial assets treated as a signal processing and optimization problem. It explores such connections and capitalizes on the existing mathematical tools developed in wireless communications and signal processing to solve real-life problems arising in the financial markets in an unprecedented way. A Signal Processing Perspective of Financial Engineering provides straightforward and systematic access to financial engineering for researchers in signal processing and communications so that they can understand problems in financial engineering more easily and may even apply signal processing techniques to handle some financial problems.","PeriodicalId":12340,"journal":{"name":"Found. Trends Signal Process.","volume":"31 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2016-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79044937","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As a major breakthrough in artificial intelligence, deep learning has achieved very impressive success in solving grand challenges in many fields including speech recognition, natural language processing, computer vision, image and video processing, and multimedia. This article provides a historical overview of deep learning and focus on its applications in object recognition, detection, and segmentation, which are key challenges of computer vision and have numerous applications to images and videos. The discussed research topics on object recognition include image classification on ImageNet, face recognition, and video classification. The detection part covers general object detection on ImageNet, pedestrian detection, face landmark detection face alignment, and human landmark detection pose estimation. On the segmentation side, thearticle discusses the most recent progress on scene labeling, semantic segmentation, face parsing, human parsing and saliency detection. Object recognition is considered as whole-image classification, while detection and segmentation are pixelwise classification tasks. Their fundamental differences will be discussed in this article. Fully convolutional neural networks and highly efficient forward and backward propagation algorithms specially designed for pixelwise classification task will be introduced. The covered application domains are also much diversified. Human and face images have regular structures, while general object and scene images have much more complex variations in geometric structures and layout. Videos include the temporal dimension. Therefore, they need to be processed with different deep models. All the selected domain applications have received tremendous attentions in the computer vision and multimedia communities. Through concrete examples of these applications, we explain the key points which make deep learning outperform conventional computer vision systems. 1 Different than traditional pattern recognition systems, which heavily rely on manually designed features, deep learning automatically learns hierarchical feature representations from massive training data and disentangles hidden factors of input data through multi-level nonlinear mappings. 2 Different than existing pattern recognition systems which sequentially design or train their key components, deep learning is able to jointly optimize all the components and crate synergy through close interactions among them. 3 While most machine learning models can be approximated with neural networks with shallow structures, for some tasks, the expressive power of deep models increases exponentially as their architectures go deep. Deep models are especially good at learning global contextual feature representation with their deep structures. 4 Benefitting from the large learning capacity of deep models, some classical computer vision challenges can be recast as high-dimensional data transform problems and can be solved from new perspectives. Final
{"title":"Deep Learning in Object Recognition, Detection, and Segmentation","authors":"Xiaogang Wang","doi":"10.1561/2000000071","DOIUrl":"https://doi.org/10.1561/2000000071","url":null,"abstract":"As a major breakthrough in artificial intelligence, deep learning has achieved very impressive success in solving grand challenges in many fields including speech recognition, natural language processing, computer vision, image and video processing, and multimedia. This article provides a historical overview of deep learning and focus on its applications in object recognition, detection, and segmentation, which are key challenges of computer vision and have numerous applications to images and videos. The discussed research topics on object recognition include image classification on ImageNet, face recognition, and video classification. The detection part covers general object detection on ImageNet, pedestrian detection, face landmark detection face alignment, and human landmark detection pose estimation. On the segmentation side, thearticle discusses the most recent progress on scene labeling, semantic segmentation, face parsing, human parsing and saliency detection. Object recognition is considered as whole-image classification, while detection and segmentation are pixelwise classification tasks. Their fundamental differences will be discussed in this article. Fully convolutional neural networks and highly efficient forward and backward propagation algorithms specially designed for pixelwise classification task will be introduced. The covered application domains are also much diversified. Human and face images have regular structures, while general object and scene images have much more complex variations in geometric structures and layout. Videos include the temporal dimension. Therefore, they need to be processed with different deep models. All the selected domain applications have received tremendous attentions in the computer vision and multimedia communities. Through concrete examples of these applications, we explain the key points which make deep learning outperform conventional computer vision systems. 1 Different than traditional pattern recognition systems, which heavily rely on manually designed features, deep learning automatically learns hierarchical feature representations from massive training data and disentangles hidden factors of input data through multi-level nonlinear mappings. 2 Different than existing pattern recognition systems which sequentially design or train their key components, deep learning is able to jointly optimize all the components and crate synergy through close interactions among them. 3 While most machine learning models can be approximated with neural networks with shallow structures, for some tasks, the expressive power of deep models increases exponentially as their architectures go deep. Deep models are especially good at learning global contextual feature representation with their deep structures. 4 Benefitting from the large learning capacity of deep models, some classical computer vision challenges can be recast as high-dimensional data transform problems and can be solved from new perspectives. Final","PeriodicalId":12340,"journal":{"name":"Found. Trends Signal Process.","volume":"14 1","pages":"217-382"},"PeriodicalIF":0.0,"publicationDate":"2016-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89327700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：