Pub Date : 2024-06-26DOI: 10.1109/OJSP.2024.3419569
AprilPyone MaungMaung;Isao Echizen;Hitoshi Kiya
In this paper, we propose key-based defense model proliferation by leveraging pre-trained models and utilizing recent efficient fine-tuning techniques on ImageNet-1 k classification. First, we stress that deploying key-based models on edge devices is feasible with the latest model deployment advancements, such as Apple CoreML, although the mainstream enterprise edge artificial intelligence (Edge AI) has been focused on the Cloud. Then, we point out that the previous key-based defense on on-device image classification is impractical for two reasons: (1) training many classifiers from scratch is not feasible, and (2) key-based defenses still need to be thoroughly tested on large datasets like ImageNet. To this end, we propose to leverage pre-trained models and utilize efficient fine-tuning techniques to proliferate key-based models even on limited compute resources. Experiments were carried out on the ImageNet-1 k dataset using adaptive and non-adaptive attacks. The results show that our proposed fine-tuned key-based models achieve a superior classification accuracy (more than 10% increase) compared to the previous key-based models on classifying clean and adversarial examples.
在本文中，我们提出了基于密钥的防御模型扩散方案，即利用预训练模型和最近在 ImageNet-1 k 分类上采用的高效微调技术。首先，我们强调，虽然主流的企业边缘人工智能（Edge AI）都集中在云端，但随着苹果 CoreML 等最新模型部署技术的发展，在边缘设备上部署基于密钥的模型是可行的。然后，我们指出，之前基于密钥的设备上图像分类防御是不切实际的，原因有二：（1）从头开始训练许多分类器是不可行的；（2）基于密钥的防御仍需在大型数据集（如 ImageNet）上进行彻底测试。为此，我们建议利用预先训练好的模型，并利用高效的微调技术，即使在有限的计算资源上也能推广基于密钥的模型。我们使用自适应和非自适应攻击在 ImageNet-1 k 数据集上进行了实验。结果表明，与以前的基于密钥的模型相比，我们提出的基于密钥的微调模型在对干净和对抗性示例进行分类时实现了更高的分类准确率（提高 10%以上）。
{"title":"Efficient Key-Based Adversarial Defense for ImageNet by Using Pre-Trained Models","authors":"AprilPyone MaungMaung;Isao Echizen;Hitoshi Kiya","doi":"10.1109/OJSP.2024.3419569","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3419569","url":null,"abstract":"In this paper, we propose key-based defense model proliferation by leveraging pre-trained models and utilizing recent efficient fine-tuning techniques on ImageNet-1 k classification. First, we stress that deploying key-based models on edge devices is feasible with the latest model deployment advancements, such as Apple CoreML, although the mainstream enterprise edge artificial intelligence (Edge AI) has been focused on the Cloud. Then, we point out that the previous key-based defense on on-device image classification is impractical for two reasons: (1) training many classifiers from scratch is not feasible, and (2) key-based defenses still need to be thoroughly tested on large datasets like ImageNet. To this end, we propose to leverage pre-trained models and utilize efficient fine-tuning techniques to proliferate key-based models even on limited compute resources. Experiments were carried out on the ImageNet-1 k dataset using adaptive and non-adaptive attacks. The results show that our proposed fine-tuned key-based models achieve a superior classification accuracy (more than 10% increase) compared to the previous key-based models on classifying clean and adversarial examples.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"902-913"},"PeriodicalIF":2.9,"publicationDate":"2024-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10572223","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141964708","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pathloss quantifies the reduction in power density of a signal radiated from a transmitter. The attenuation is due to large-scale effects such as free-space propagation loss and interactions (e.g., penetration, reflection, and diffraction) of the signal with objects such as buildings, vehicles, trees, and pedestrians in the propagation environment. Many current or planned wireless communications applications require the knowledge (or a reliable approximation) of the pathloss on a dense grid (radio map) of the environment of interest. Deterministic simulation methods such as ray tracing are known to provide very good estimates of pathloss values. However, their high computational complexity makes them unsuitable for most of the applications envisaged. To promote research and facilitate a fair comparison among the recently proposed fast and accurate deep learning-based pathloss radio map prediction methods, we have organized the ICASSP 2023 First Pathloss Radio Map Prediction Challenge. In this overview paper, we describe the pathloss radio map prediction problem, provide a literature survey of the current state of the art, describe the challenge datasets, the challenge task, and the challenge evaluation methodology. Finally, we provide a brief overview of the submitted methods and present the results of the challenge.
{"title":"Overview of the First Pathloss Radio Map Prediction Challenge","authors":"Çağkan Yapar;Fabian Jaensch;Ron Levie;Gitta Kutyniok;Giuseppe Caire","doi":"10.1109/OJSP.2024.3419563","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3419563","url":null,"abstract":"Pathloss quantifies the reduction in power density of a signal radiated from a transmitter. The attenuation is due to large-scale effects such as free-space propagation loss and interactions (e.g., penetration, reflection, and diffraction) of the signal with objects such as buildings, vehicles, trees, and pedestrians in the propagation environment. Many current or planned wireless communications applications require the knowledge (or a reliable approximation) of the pathloss on a dense grid (radio map) of the environment of interest. Deterministic simulation methods such as ray tracing are known to provide very good estimates of pathloss values. However, their high computational complexity makes them unsuitable for most of the applications envisaged. To promote research and facilitate a fair comparison among the recently proposed fast and accurate deep learning-based pathloss radio map prediction methods, we have organized the ICASSP 2023 First Pathloss Radio Map Prediction Challenge. In this overview paper, we describe the pathloss radio map prediction problem, provide a literature survey of the current state of the art, describe the challenge datasets, the challenge task, and the challenge evaluation methodology. Finally, we provide a brief overview of the submitted methods and present the results of the challenge.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"948-963"},"PeriodicalIF":2.9,"publicationDate":"2024-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10572278","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141993995","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-26DOI: 10.1109/OJSP.2024.3419448
Fausto García-Gangoso;Fernando Cruz-Roldán
The use of an accurate cyclic prefix length is crucial in orthogonal frequency division multiplexing (OFDM) to avoid intercarrier and intersymbol interference. Although there have been many works that analyse the interference of windowed OFDM, this study remains open in the context of power line communications (PLCs) taking into account the physical-layer (PHY) specifications of the standards. This paper focuses on obtaining a closed-form expression of the input-output relationship in windowed OFDM power line communication (PLC) systems under the condition of insufficient cyclic prefix, while incorporating various blocks deployed in the PHY under IEEE 1901 standards. The derived analysis is important for quantifying the undesired signal component in each subcarrier at a specific time, which renders the detection of the corresponding symbol more difficult. Moreover, a novel procedure is proposed that allows the use of a smaller number of redundant samples to avoid interference. This novel procedure, performed in the receiver after the windowing stage, replaces the overlap-and-add operations with multiplications, offering the advantage of requiring fewer samples from the time-domain received signal to recover each transmitted data symbol. Numerical results demonstrate the feasibility of interference-free transmission on channels with a larger number of samples, thereby yielding better results across various PLC scenarios.
{"title":"SINR Analysis of Windowed OFDM in Power Line Communication Systems","authors":"Fausto García-Gangoso;Fernando Cruz-Roldán","doi":"10.1109/OJSP.2024.3419448","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3419448","url":null,"abstract":"The use of an accurate cyclic prefix length is crucial in orthogonal frequency division multiplexing (OFDM) to avoid intercarrier and intersymbol interference. Although there have been many works that analyse the interference of windowed OFDM, this study remains open in the context of power line communications (PLCs) taking into account the physical-layer (PHY) specifications of the standards. This paper focuses on obtaining a closed-form expression of the input-output relationship in windowed OFDM power line communication (PLC) systems under the condition of insufficient cyclic prefix, while incorporating various blocks deployed in the PHY under IEEE 1901 standards. The derived analysis is important for quantifying the undesired signal component in each subcarrier at a specific time, which renders the detection of the corresponding symbol more difficult. Moreover, a novel procedure is proposed that allows the use of a smaller number of redundant samples to avoid interference. This novel procedure, performed in the receiver after the windowing stage, replaces the overlap-and-add operations with multiplications, offering the advantage of requiring fewer samples from the time-domain received signal to recover each transmitted data symbol. Numerical results demonstrate the feasibility of interference-free transmission on channels with a larger number of samples, thereby yielding better results across various PLC scenarios.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"1052-1060"},"PeriodicalIF":2.9,"publicationDate":"2024-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10572224","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142518177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-24DOI: 10.1109/OJSP.2024.3397168
Alexander Bertrand;Ozlem Kalinli
The 2023 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2023) took place in Rhodos, Greece, running from June 4th to June 10th, with a record number of paper submissions and attendees. Since 2021, ICASSP has featured the “Signal Processing Grand Challenges” (SPGC) program, which has become an annual highlight at the conference. ICASSP 2023 featured a record number of 15 SPGCs, carefully selected from a large number of submissions, and covering a wide variety of application domains, including audio, acoustics, speech, biomedical signals, communications, and image processing. A list of accepted SPGCs can be found at https://2023.ieeeicassp.org/signal-processing-grand-challenges/ , which also includes links to detailed information for each challenge.
{"title":"Editorial: Special Issue on the ICASSP 2023 Signal Processing Grand Challenges","authors":"Alexander Bertrand;Ozlem Kalinli","doi":"10.1109/OJSP.2024.3397168","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3397168","url":null,"abstract":"The 2023 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2023) took place in Rhodos, Greece, running from June 4th to June 10th, with a record number of paper submissions and attendees. Since 2021, ICASSP has featured the “Signal Processing Grand Challenges” (SPGC) program, which has become an annual highlight at the conference. ICASSP 2023 featured a record number of 15 SPGCs, carefully selected from a large number of submissions, and covering a wide variety of application domains, including audio, acoustics, speech, biomedical signals, communications, and image processing. A list of accepted SPGCs can be found at \u0000<uri>https://2023.ieeeicassp.org/signal-processing-grand-challenges/</uri>\u0000, which also includes links to detailed information for each challenge.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"630-631"},"PeriodicalIF":2.9,"publicationDate":"2024-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10570160","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141447970","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-14DOI: 10.1109/OJSP.2024.3402154
Elizabeth Coda;Brad Clymer;Chance DeSmet;Yijing Watkins;Michael Girard
A wide variety of adversarial attacks have been proposed and explored using image and audio data. These attacks are notoriously easy to generate digitally when the attacker can directly manipulate the input to a model, but are much more difficult to implement in the real world. In this paper we present a universal, time invariant attack for general time series data such that the attack has a frequency spectrum primarily composed of the frequencies present in the original data. The universality of the attack makes it fast and easy to implement as no computation is required to add it to an input, while time invariance is useful for real world deployment. Additionally, the frequency constraint ensures the attack can withstand filtering defenses. We demonstrate the effectiveness of the attack on two different classification tasks through both digital and real world experiments, and show that the attack is robust against common transform-and-compare defense pipelines.
{"title":"Universal Fourier Attack for Time Series","authors":"Elizabeth Coda;Brad Clymer;Chance DeSmet;Yijing Watkins;Michael Girard","doi":"10.1109/OJSP.2024.3402154","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3402154","url":null,"abstract":"A wide variety of adversarial attacks have been proposed and explored using image and audio data. These attacks are notoriously easy to generate digitally when the attacker can directly manipulate the input to a model, but are much more difficult to implement in the real world. In this paper we present a universal, time invariant attack for general time series data such that the attack has a frequency spectrum primarily composed of the frequencies present in the original data. The universality of the attack makes it fast and easy to implement as no computation is required to add it to an input, while time invariance is useful for real world deployment. Additionally, the frequency constraint ensures the attack can withstand filtering defenses. We demonstrate the effectiveness of the attack on two different classification tasks through both digital and real world experiments, and show that the attack is robust against common transform-and-compare defense pipelines.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"858-866"},"PeriodicalIF":2.9,"publicationDate":"2024-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10557789","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141725683","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-07DOI: 10.1109/OJSP.2024.3411468
Augusto Aubry;Marco Boddi;Antonio De Maio;Massimo Rosamilia
This paper addresses the Direction-of-Arrival (DOA) estimation problem using a narrowband polarimetric array sensing system. The considered receiving equipment is composed of two sub-arrays of sensors with orthogonal polarizations. By suitably modeling the received signal via a sparse representation (accounting for the multiple snapshots and the polarimetric array manifold structure), two iterative algorithms, namely Polarimetric Sparse Learning via Iterative Minimization (POL-SLIM) and Polarimetric Sparse Iterative Covariance-based Estimation (POL-SPICE), are devised to accomplish the estimation task. The proposed algorithms provide accurate DOA estimates while enjoying nice (rigorously proven) convergence properties. Numerical analysis shows the effectiveness of POL-SLIM and POL-SPICE to successfully locate signal sources in both passive sensing applications (with large numbers of collected snapshots) and radar spatial processing, also in comparison with single-polarization counterparts as well as theoretical benchmarks.
本文利用窄带偏振阵列传感系统解决到达方向（DOA）估计问题。所考虑的接收设备由两个具有正交极化的传感器子阵列组成。通过稀疏表示对接收信号进行适当建模（考虑到多个快照和偏振阵列流形结构），设计了两种迭代算法，即通过迭代最小化进行偏振稀疏学习（POL-SLIM）和基于协方差的偏振稀疏迭代估计（POL-SPICE），以完成估计任务。所提出的算法可提供精确的 DOA 估计值，同时具有良好的（经严格证明的）收敛特性。数值分析表明，POL-SLIM 和 POL-SPICE 在被动传感应用（收集大量快照）和雷达空间处理中成功定位信号源的有效性，与单极化对应算法和理论基准相比也是如此。
{"title":"Sparse DOA Estimation With Polarimetric Arrays","authors":"Augusto Aubry;Marco Boddi;Antonio De Maio;Massimo Rosamilia","doi":"10.1109/OJSP.2024.3411468","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3411468","url":null,"abstract":"This paper addresses the Direction-of-Arrival (DOA) estimation problem using a narrowband polarimetric array sensing system. The considered receiving equipment is composed of two sub-arrays of sensors with orthogonal polarizations. By suitably modeling the received signal via a sparse representation (accounting for the multiple snapshots and the polarimetric array manifold structure), two iterative algorithms, namely Polarimetric Sparse Learning via Iterative Minimization (POL-SLIM) and Polarimetric Sparse Iterative Covariance-based Estimation (POL-SPICE), are devised to accomplish the estimation task. The proposed algorithms provide accurate DOA estimates while enjoying nice (rigorously proven) convergence properties. Numerical analysis shows the effectiveness of POL-SLIM and POL-SPICE to successfully locate signal sources in both passive sensing applications (with large numbers of collected snapshots) and radar spatial processing, also in comparison with single-polarization counterparts as well as theoretical benchmarks.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"886-901"},"PeriodicalIF":2.9,"publicationDate":"2024-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10552043","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141725564","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-05DOI: 10.1109/OJSP.2024.3410089
Martin Jälmby;Filip Elvander;Toon van Waterschoot
The room impulse response (RIR) describes the response of a room to an acoustic excitation signal and models the acoustic channel between a point source and receiver. RIRs are used in a wide range of applications, e.g., virtual reality. In such an application, the availability of closely spaced RIRs and the capability to achieve low latency are imperative to provide an immersive experience. However, representing a complete acoustic environment using a fine grid of RIRs is prohibitive from a storage point of view and without exploiting spatial proximity, acoustic rendering becomes computationally expensive. We therefore propose two methods for the joint compression of multiple RIRs, based on the generalized low-rank approximation of matrices (GLRAM), for the purpose of efficiently storing RIRs and allowing for low-latency convolution. We show how one of the components of the GLRAM decomposition is virtually invariant to the change of position of the source throughout the room and how this can be exploited in the modeling and convolution. In simulations we show how this offers high compression, with less quality degradation than comparable benchmark methods.
{"title":"Multi-Channel Low-Rank Convolution of Jointly Compressed Room Impulse Responses","authors":"Martin Jälmby;Filip Elvander;Toon van Waterschoot","doi":"10.1109/OJSP.2024.3410089","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3410089","url":null,"abstract":"The room impulse response (RIR) describes the response of a room to an acoustic excitation signal and models the acoustic channel between a point source and receiver. RIRs are used in a wide range of applications, e.g., virtual reality. In such an application, the availability of closely spaced RIRs and the capability to achieve low latency are imperative to provide an immersive experience. However, representing a complete acoustic environment using a fine grid of RIRs is prohibitive from a storage point of view and without exploiting spatial proximity, acoustic rendering becomes computationally expensive. We therefore propose two methods for the joint compression of multiple RIRs, based on the generalized low-rank approximation of matrices (GLRAM), for the purpose of efficiently storing RIRs and allowing for low-latency convolution. We show how one of the components of the GLRAM decomposition is virtually invariant to the change of position of the source throughout the room and how this can be exploited in the modeling and convolution. In simulations we show how this offers high compression, with less quality degradation than comparable benchmark methods.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"850-857"},"PeriodicalIF":2.9,"publicationDate":"2024-06-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10549780","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141543936","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-04-30DOI: 10.1109/OJSP.2024.3395179
Yuval Becker;Raz Z. Nossek;Tomer Peleg
Image demosaicing is an important step in the image processing pipeline for digital cameras. In data centric approaches, such as deep learning, the distribution of the dataset used for training can impose a bias on the networks' outcome. For example, in natural images most patches are smooth, and high-content patches are much rarer. This can lead to a bias in the performance of demosaicing algorithms. Most deep learning approaches address this challenge by utilizing specific losses or designing special network architectures. We propose a novel approach SDAT , Sub-Dataset Alternation Training, that tackles the problem from a training protocol perspective. SDAT is comprised of two essential phases. In the initial phase, we employ a method to create sub-datasets from the entire dataset, each inducing a distinct bias. The subsequent phase involves an alternating training process, which uses the derived sub-datasets in addition to training also on the entire dataset. SDAT can be applied regardless of the chosen architecture as demonstrated by various experiments we conducted for the demosaicing task. The experiments are performed across a range of architecture sizes and types, namely CNNs and transformers. We show improved performance in all cases. We are also able to achieve state-of-the-art results on three highly popular image demosaicing benchmarks.
{"title":"SDAT: Sub-Dataset Alternation Training for Improved Image Demosaicing","authors":"Yuval Becker;Raz Z. Nossek;Tomer Peleg","doi":"10.1109/OJSP.2024.3395179","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3395179","url":null,"abstract":"Image demosaicing is an important step in the image processing pipeline for digital cameras. In data centric approaches, such as deep learning, the distribution of the dataset used for training can impose a bias on the networks' outcome. For example, in natural images most patches are smooth, and high-content patches are much rarer. This can lead to a bias in the performance of demosaicing algorithms. Most deep learning approaches address this challenge by utilizing specific losses or designing special network architectures. We propose a novel approach \u0000<bold>SDAT</b>\u0000, Sub-Dataset Alternation Training, that tackles the problem from a training protocol perspective. SDAT is comprised of two essential phases. In the initial phase, we employ a method to create sub-datasets from the entire dataset, each inducing a distinct bias. The subsequent phase involves an alternating training process, which uses the derived sub-datasets in addition to training also on the entire dataset. SDAT can be applied regardless of the chosen architecture as demonstrated by various experiments we conducted for the demosaicing task. The experiments are performed across a range of architecture sizes and types, namely CNNs and transformers. We show improved performance in all cases. We are also able to achieve state-of-the-art results on three highly popular image demosaicing benchmarks.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"611-620"},"PeriodicalIF":0.0,"publicationDate":"2024-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10510581","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140919094","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-04-26DOI: 10.1109/OJSP.2024.3394369
Jonathan Shani;Tom Tirer;Raja Giryes;Tamir Bendory
We study the 2D super-resolution multi-reference alignment (SR-MRA) problem: estimating an image from its down-sampled, circularly translated, and noisy copies. The SR-MRA problem serves as a mathematical abstraction of the structure determination problem for biological molecules. Since the SR-MRA problem is ill-posed without prior knowledge, accurate image estimation relies on designing priors that describe the statistics of the images of interest. In this work, we build on recent advances in image processing and harness the power of denoisers as priors for images. To estimate an image, we propose utilizing denoisers as projections and using them within two computational frameworks that we propose: projected expectation-maximization and projected method of moments. We provide an efficient GPU implementation and demonstrate the effectiveness of these algorithms through extensive numerical experiments on a wide range of parameters and images.
{"title":"Denoiser-Based Projections for 2D Super-Resolution MRA","authors":"Jonathan Shani;Tom Tirer;Raja Giryes;Tamir Bendory","doi":"10.1109/OJSP.2024.3394369","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3394369","url":null,"abstract":"We study the 2D super-resolution multi-reference alignment (SR-MRA) problem: estimating an image from its down-sampled, circularly translated, and noisy copies. The SR-MRA problem serves as a mathematical abstraction of the structure determination problem for biological molecules. Since the SR-MRA problem is ill-posed without prior knowledge, accurate image estimation relies on designing priors that describe the statistics of the images of interest. In this work, we build on recent advances in image processing and harness the power of denoisers as priors for images. To estimate an image, we propose utilizing denoisers as projections and using them within two computational frameworks that we propose: projected expectation-maximization and projected method of moments. We provide an efficient GPU implementation and demonstrate the effectiveness of these algorithms through extensive numerical experiments on a wide range of parameters and images.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"621-629"},"PeriodicalIF":0.0,"publicationDate":"2024-04-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10508939","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141078833","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-04-16DOI: 10.1109/OJSP.2024.3389825
Marc García-Bermúdez;Jordi Solé-Lloveras;Martin Hudlička;Marco A. Azpúrua
The standard CISPR 16-1-1 defines the measuring receiver using a black-box approach and sets requirements for its accuracy and spectral properties. Traditionally, such test receivers were developed using a superheterodyne architecture. Recently, time-domain electromagnetic emission measurement systems have been built employing direct sampling instruments, mainly oscilloscopes, and relying on specific signal processing to emulate the performance of compliant instruments. In these cases, the short-time Fourier transform is used for spectral estimation, but the corresponding electromagnetic compatibility standards lack details for its correct use with respect to parameters such as windowing function, overlapping factor, and frequency interpolation. Moreover, it is unclear which combination of spectral estimation parameters is best fit for this purpose. Obtaining reliable, consistent and low uncertainty spectral estimates of electromagnetic emissions measured in time-domain needs appropriate configuration and tuning of the signal processing algorithms. This paper investigates the error in the calculated spectrum for various reference signals: multitone, chirp pulses and rectangular pulses. The analysis is carried out for each CISPR band from A to D, that is, between 9 kHz and 1 GHz. After $489.6times 10^{3}$ iterations, distributed in 1700 different digital implementations of the CISPR 16-1-1 measuring receiver, the simulations outcomes point to certain sets of parameters that showed satisfactory performance overall, being the Nutall, Kaiser, and Parzen windows with more than 75% of overlapping and using interpolation factor higher than 5, generally suitable. Calibration results are used to experimentally verify that a valid set of parameters is adequate to fulfil CISPR 16-1-1 requirements.
{"title":"Evaluation of Spectral Estimation Parameters for Direct Sampling FFT-Based Measuring Receivers","authors":"Marc García-Bermúdez;Jordi Solé-Lloveras;Martin Hudlička;Marco A. Azpúrua","doi":"10.1109/OJSP.2024.3389825","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3389825","url":null,"abstract":"The standard CISPR 16-1-1 defines the measuring receiver using a black-box approach and sets requirements for its accuracy and spectral properties. Traditionally, such test receivers were developed using a superheterodyne architecture. Recently, time-domain electromagnetic emission measurement systems have been built employing direct sampling instruments, mainly oscilloscopes, and relying on specific signal processing to emulate the performance of compliant instruments. In these cases, the short-time Fourier transform is used for spectral estimation, but the corresponding electromagnetic compatibility standards lack details for its correct use with respect to parameters such as windowing function, overlapping factor, and frequency interpolation. Moreover, it is unclear which combination of spectral estimation parameters is best fit for this purpose. Obtaining reliable, consistent and low uncertainty spectral estimates of electromagnetic emissions measured in time-domain needs appropriate configuration and tuning of the signal processing algorithms. This paper investigates the error in the calculated spectrum for various reference signals: multitone, chirp pulses and rectangular pulses. The analysis is carried out for each CISPR band from A to D, that is, between 9 kHz and 1 GHz. After \u0000<inline-formula><tex-math>$489.6times 10^{3}$</tex-math></inline-formula>\u0000 iterations, distributed in 1700 different digital implementations of the CISPR 16-1-1 measuring receiver, the simulations outcomes point to certain sets of parameters that showed satisfactory performance overall, being the Nutall, Kaiser, and Parzen windows with more than 75% of overlapping and using interpolation factor higher than 5, generally suitable. Calibration results are used to experimentally verify that a valid set of parameters is adequate to fulfil CISPR 16-1-1 requirements.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"588-598"},"PeriodicalIF":0.0,"publicationDate":"2024-04-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10502171","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140818796","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","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：