检索结果-内蒙古大学图书馆

Unsupervised Low-Light Image Enhancement by Extracting Structural Similarity and Color Consistency

IEEE signal processing LETTERS 2022年 29卷 997-1001页

作者： Shi, Yangming Wang, Binquan Wu, Xiaopo Zhu, Ming Univ Sci & Technol China Dept Elect Engn Hefei 230027 Anhui Peoples R China

A novel structure-aware unsupervised network is proposed to deal with low-light image enhancement issues based on the inspiration of Retinex theory and self-supervised perceptual loss. It comprises four main components, namely the original structural similarity module, the novel color consistency module, the attentional enhancement module, and the naturalness discriminator module. Specially for the structural similarity module, an embedded structural feature extractor (SFE) model capable of generating structure correspondence is well designed and pre-trained by employing the contrastive learning technique, and a multi-scale structural similarity distance is introduced to optimize the SFE network. Besides, a self-supervised color consistency module is established by using a degraded estimation algorithm for recovering the missing colors. The whole enhancement framework operates in unsupervised manners and finally obtains the best naturalness image quality evaluator metric. Experimental results demonstrate that the proposed unsupervised network is able to recover natural structure and color images more effectively, which would also help to enlarge the practical application without collecting paired datasets in advance.

关键词： Image color analysis Feature extraction Training Optimization signal processing algorithms Brightness Lighting Color consistency generative adversarial network (GAN) structural similarity unsupervised learning

来源：评论

学校读者我要写书评

暂无评论

RSD-GAN: Regularized Sobolev Defense GAN Against Speech-to-Text Adversarial Attacks

引用

IEEE signal processing LETTERS 2022年 29卷 1998-2002页

作者： Esmaeilpour, Mohammad Chaalia, Nourhene Cardinal, Patrick Univ Quebec Ecole Technol Super ETS Montreal PQ H2L 2C4 Canada Federat Caisses Desjardins Quebec Montreal PQ H2X 3Y2 Canada

This letter introduces a new synthesis-based defense algorithm for counteracting with a varieties of adversarial attacks developed for challenging the performance of the cutting-edge speech-to-text transcription systems. Our algorithm implements a Sobolev-based GAN and proposes a novel regularizer for effectively controlling over the functionality of the entire generative model, particularly the discriminator network during training. Our achieved results upon carrying out numerous experiments on the victim DeepSpeech, Kaldi, and Lingvo speech transcription systems corroborate the remarkable performance of our defense approach against a comprehensive range of targeted and non-targeted adversarial attacks.

关键词： Generative adversarial networks Training Perturbation methods signal processing algorithms Generators Optimization Psychoacoustic models Adversarial defense GAN regularization Sobolev integral probability metric speech adversarial attack

来源：评论

学校读者我要写书评

暂无评论

EnhanceFace: Adaptive Weighted SoftMax Loss for Deep Face Recognition

引用

IEEE signal processing LETTERS 2022年 29卷 65-69页

作者： Wang, Jiaqi Zheng, Chen Yang, Xiaohui Yang, Lijun Henan Univ Sch Math & Stat Kaifeng 475004 Peoples R China Henan Univ Henan Engn Res Ctr Artificial Intelligence Theory Kaifeng 475004 Peoples R China

Loss function is an important topic in the field of face recognition, while the margin-based loss function is one of the most useful methods to enhance discriminability. Recently, the method of dividing the samples into easy and hard ones effectively optimizes the margin-based loss function by emphasizing these two parts at different training stages. However, the hard samples contain not only intra-personal variations that can be easily misclassified but also wrong-labeled samples. To capture and distinguish their difference, an EnhanceFace method is proposed with a new adaptive weighted loss function in this letter to improve the accuracy of face recognition. Firstly, this method partitions the hard samples into semi-hard and harder samples. Then, a new loss function is developed to adaptively adjust corresponding weights for both semi-hard samples and harder samples during the training process. With this new loss function, their effects can be well balanced according to their importance. The EnhanceFace method is tested on different popular benchmarks. Finally, better performance is shown by the experimental results comparing with other state-of-the-art methods.

关键词： Training Face recognition Loss measurement Weight measurement signal processing algorithms Optimization Feature extraction Face recognition loss function adaptive weight

来源：评论

学校读者我要写书评

暂无评论

Large-Scale Multi-View Clustering via Fast Essential Subspace Representation Learning

引用

IEEE signal processing LETTERS 2022年 29卷 1893-1897页

作者： Zheng, Qinghai Fuzhou Univ Coll Comp & Data Sci Fuzhou 350108 Peoples R China

Large-scale Multi-View Clustering (LMVC) is a hot research problem in the fields of signal processing and machine learning, and many anchor-based multi-view subspace clustering algorithms are proposed in recent years. However, most existing methods usually concentrate on the issue of reducing the time cost and ignore the exploration of the complementary information during the clustering process. To this end, we propose a Fast Essential Subspace Representation Learning (FESRL) method for large-scale multi-view subspace clustering. Specifically, FESRL introduces the orthogonal transformation to investigate both the complementary and consensus information across multiple views. The essential subspace representation can be learned in a linear time cost. Experiments conducted on several benchmark datasets illustrate the competitiveness of the proposed method.

关键词： signal processing algorithms Optimization Costs Representation learning Time complexity Clustering algorithms Convergence Large-scale multi-view subspace clustering linear computational complexity

来源：评论

学校读者我要写书评

暂无评论

Efficient Design of Scaled Rectangular (Saramaki) Window

引用

IEEE signal processing LETTERS 2022年 29卷 489-492页

作者： Lim, Yong Ching Liu, Qinglai Diniz, Paulo S. R. Saramaki, Tapio DASP Pte Ltd Singapore Singapore Tampere Univ Tampere 33101 Finland Univ Fed Rio de Janeiro BR-21941972 Rio De Janeiro Brazil Nanyang Technol Univ Temasek Labs Singapore 637553 Singapore

A rectangular-window sidelobe magnitude reduction method by widening the main lobe width was proposed by Saramaki. A technique for trading off main lobe width against sidelobe magnitude for any arbitrary window was reported by Lim et al.;a fast convergence algorithm for its implementation was proposed by the same authors in another article, where the derivatives of the window function are expressed in Chebyshev polynomials which have high arithmetic complexity. All the coefficients of a rectangular-window are equal;this special property is exploited, in this paper, for deriving the window function's derivatives without the use of Chebyshev polynomials resulting in a great reduction in the arithmetic complexity.

关键词： Arithmetic Complexity theory Chebyshev approximation signal processing algorithms Real-time systems Convergence Laboratories Scaled rectangular (Saramaki) window sidelobe magnitude main lobe width main lobe sidelobe tradeoff

来源：评论

学校读者我要写书评

暂无评论

Code Optimization and Angle-Doppler Imaging for ST-CDM LFMCW MIMO Radar Systems

引用

IEEE signal processing LETTERS 2022年 29卷 2043-2047页

作者： Shang, Xiaolei Cheng, Yuanbo Univ Sci & Technol China Dept Elect Engn & Informat Sci Hefei 230027 Peoples R China

We consider code optimization and angle-Doppler imaging for slow-time code division multiplexing (ST-CDM) linear frequency-modulated continuous-wave (LFMCW) multiple-input multiple-output (MIMO) radar systems. We optimize the slow-time code via the minimization of a Cramer-Rao Bound (CRB)-based metric to enhance the parameter estimation performance. Then, a computationally efficient RELAX-based algorithm is presented to obtain the maximum likelihood (ML) estimates of the target angle-Doppler parameters. Numerical examples show that the proposed approaches can be used to improve the performance of parameter estimation and angle-Doppler imaging of ST-CDM LFMCW MIMO radar systems.

关键词： Codes Radio frequency MIMO radar Maximum likelihood estimation Radar Radar imaging signal processing algorithms Angle-Doppler imaging Cramer-Rao bound (CRB) LFMCW maximum-likelihood estimation MIMO radar RELAX slow-time code optimization ST-CDM

来源：评论

学校读者我要写书评

暂无评论

Modified Multi-Direction Iterative Algorithm for Separable Nonlinear Models With Missing Data

引用

IEEE signal processing LETTERS 2022年 29卷 1968-1972页

作者： Chen, Jing Hu, Manfeng Mao, Yawen Zhu, Quanmin Jiangnan Univ Sch Sci Wuxi 214122 Jiangsu Peoples R China Univ West England Dept Engn Design & Math Bristol BS16 1QY Avon England

Multi-direction iterative (MUL-DI) algorithm is an efficient algorithm for large-scale models, and it establishes a theoretical linkage between least squares (LS) and gradient descent (GD) algorithms. However, it involves Givens transformation and dense matrix calculation in each iteration, which leads to heavy computational efforts. In this letter, a modified MUL-DI algorithm is proposed for separable nonlinear models with missing data. Several directions are designed using a diagonal matrix, and their corresponding step-sizes are obtained based on LS algorithm. Compared with the traditional algorithms, the algorithm proposed in this letter has the following advantages: (1) has a faster convergence rate;(2) has a simple cost function;(3) is more robust to the condition number;(4) has less computational efforts. A simulation example shows the effectiveness of the modified MUL-DI algorithm.

关键词： signal processing algorithms Convergence Biological system modeling Computational modeling Data models Mathematical models Cost function Convergence rate multi-direction iterative algorithm missing data separable nonlinear model

来源：评论

学校读者我要写书评

暂无评论

An Iterative Coordinate Descent Algorithm to Compute Sparse Low-Rank Approximations

引用

IEEE signal processing LETTERS 2022年 29卷 249-253页

作者： Rusu, Cristian Univ Politehn Bucuresti Fac Automat Control & Comp Sci Bucharest 060042 Romania Univ Bucharest Fac Math & Comp Sci Dept Comp Sci Res Ctr Log Optimizat & Secur LOS Bucharest 060042 Romania

In this paper, we describe a new algorithm to build a few sparse principal components from a given data matrix. Our approach does not explicitly create the covariance matrix of the data and can be viewed as an extension of the Kogbetliantz algorithm to build an approximate singular value decomposition for a few principal components. We show the performance of the proposed algorithm to recover sparse principal components on various datasets from the literature and perform dimensionality reduction for classification applications.

关键词： signal processing algorithms Optimization Approximation algorithms Principal component analysis Covariance matrices Sparse matrices Singular value decomposition Sparse principal component analysis singular value decomposition approximation low-rank approximation

来源：评论

学校读者我要写书评

暂无评论

Error-Diffusion Based Speech Feature Quantization for Small-Footprint Keyword Spotting

引用

IEEE signal processing LETTERS 2022年 29卷 1357-1361页

作者： Luo, Mengjie Wang, Dingyi Wang, Xiaoqin Qiao, Shushan Zhou, Yumei Chinese Acad Sci Inst Microelect Beijing 100029 Peoples R China Univ Chinese Acad Sci Beijing 100029 Peoples R China

Neural network based keyword spotting (KWS) system is a critical component for user interaction in current smart devices. Although small-footprint networks have been widely explored to reduce deployment overhead, low-precision input feature representation still lacks in-depth research. In this letter, an error-diffusion based speech feature quantization method is proposed. Specifically, our algorithm adapts image processing to quantize the input speech feature maps in arbitrary bits. Experiments show that in the 10-keyword KWS task, our 3-bit representation only brings a 0.45% average accuracy drop compared to the full-precision log-Mel spectrograms while others drop over 3%. In the 2 keywords task, our 3-bit representation produces no significant differences, while 1-bit quantization only leads to an average of 1.7% accuracy drop and is even capable of handling similar keywords and imbalanced data distribution. The result proves our method, to the best of our knowledge, is the first practical method that supports as low as 1-bit quantization for single-channel speech features in small-footprint KWS. In addition, we analyze the impact of error-diffusion directions and conclude that time-direction diffusion is more suitable for temporal convolutional networks.

关键词： Quantization (signal) Task analysis Spectrogram signal processing algorithms Filter banks Standards Speech processing Keyword spotting speech feature quantization error diffusion image processing convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Rare Events via Cross-Entropy Population Monte Carlo

引用

IEEE signal processing LETTERS 2022年 29卷 439-443页

作者： Miller, Caleb Corcoran, Jem N. Schneider, Michael D. Univ Colorado Appl Math Dept Boulder CO 80309 USA Lawrence Livermore Natl Lab Astron & Astrophys Analyt Phys Div Livermore CA 94550 USA

Rare events are events that happen with very low frequency. Estimating rare event probabilities using Monte Carlo techniques is computationally expensive, often to the point of intractability, and special methods are required. Importance sampling (IS) is a well known technique that uses a proposal distribution in place of a target distribution to lower the variance of the estimator. Key to the success of IS methods is the choice of a proposal distribution, or the parameters governing the distribution. Adaptive importance sampling improves the parameters of a family or population of proposal distributions iteratively through trials. We present a novel cross-entropy population Monte Carlo algorithm, which adapts the parameters of proposals through the cross-entropy method. The proposed method stands apart from previous work in that we are not optimizing a mixture distribution. Instead, we leverage deterministic mixture weights and optimize the distributions individually through a reinterpretation of the typical derivation of the cross-entropy method. Demonstrations on rare event examples show that the algorithm can outperform existing resampling based population Monte Carlo methods, especially for higher-dimensional problems. We also demonstrate efficacy on a conjunction analysis problem.

关键词： Proposals Monte Carlo methods Statistics Sociology signal processing algorithms Artificial intelligence Optimization Adaptive importance sampling cross-entropy rare events

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：