在线更新噪声基矩阵的非负矩阵分解语音增强方法

基于非负矩阵分解(Nonnegative matrix factorization, NMF)的语音增强算法需要和背景噪声类型匹配的噪声基矩阵(Basis matrix)，而在实际中，这是很难被保证的。本文提出了一种基于噪声基矩阵在线更新的非负矩阵分解语音增强方法，该方法首先利用一个无语音帧判决模块识别出带噪语音的无语音区域，然后利用一个固定长度的滑动窗口(Sliding window)来包含若干帧最近过去的带噪语音的无语音帧，并用这些无语音帧的幅度谱在线更新噪声基矩阵，最后利用更新得到的噪声基矩阵和预先训练的语音基矩阵实现语音增强。该方法能够在线更新出匹配的噪声基矩阵，有效地解决了噪声基矩阵不匹配的问题。实验证明，本文所提的方法在线学习到的噪声基矩阵在大多数条件下比匹配训练集下训练得到的噪声基矩阵的性能还要优越。      

Online video scene clustering by competitive incremental NMF

Efficient clustering and categorizing of video are becoming more and more vital in various applications including video summarization, content-based representation and so on. The large volume of video data is the biggest challenge that this task presents, for most the clustering techniques suffer from high dimensional data in terms of both accuracy and efficiency. In addition to this, most video applications require online processing; therefore, clustering should also be done online for such tasks. This paper presents an online video scene clustering/segmentation method that is based on incremental nonnegative matrix factorization (INMF), which has been shown to be a powerful content representation tool for high dimensional data. The proposed algorithm (Comp-INMF) enables online representation of video content and increases efficiency significantly by integrating a competitive learning scheme into INMF. It brings a systematic solution to the issue of rank selection in nonnegative matrix factorization, which is equivalent to specifying the number of clusters. The clustering performance is evaluated by tests on TRECVID video sequences, and a performance comparison to baseline methods including Adaptive Resonance Theory (ART) is provided in order to demonstrate the efficiency and efficacy of the proposed video clustering scheme. Clustering performance reported in terms of recall, precision and F1 measures shows that the labeling accuracy of the algorithm is notable, especially at edit effect regions that constitute a challenging point in video analysis.     

加性噪声条件下鲁棒说话人确认

基于非负矩阵分解的语音去噪,在提高语音信号信噪比的同时,也会引起语音失真,从而导致噪声环境下说话人确认系统性能下降.本文提出基于分区约束非负矩阵分解的语音去噪方法（Nonnegative Matrix Factorization with Partial Constrains,PCNMF）,目的是在未知和非平稳噪声条件下提高话人确认系统的鲁棒性.PCNMF在满足分区约束条件的基础上分别构建语音字典和噪声字典.考虑到传统语音训练产生的语音字典往往含有一定的噪声成分,PCNMF通过数学模型产生基音及泛音频谱,在此基础上利用该频谱模仿人声的共振峰结构来合成字典,从而保证语音字典纯净性.另一方面,为了克服传统噪声字典构建方法带来的部分噪声信息丢失问题,PCNMF对在线分离出的噪声样本进行分帧和短时傅里叶变换,然后以帧为单位线性组合生成噪声字典.性能评估实验引入了多种噪声类型,实验结果表明PCNMF可有效提高说话人确认系统的鲁棒性,特别是在未知和非平稳噪声条件下其等错率相比基线系统（Multi-Condition）平均降低了5.2%.     

基于非负矩阵分解的语音增强方法综述

语音增强在语音信号处理领域举足轻重，其目的在于减少背景噪声对语音信号的影响。然而，如何从极度非平稳噪声环境下有效地分离出目标语音仍然是一个具有挑战性的问题。基于非负矩阵分解(Nonnegative matrix factorization, NMF)的语音增强算法利用非负的语音和噪声基矩阵来建模语音和噪声的频谱子空间，是目前一种先进的对抑制非平稳噪声非常有效的技术。本文首先详细地介绍了非负矩阵分解理论，包括非负矩阵分解模型，代价函数(Cost function)的定义以及常用的乘法更新准则(Multiplicative update rules)。然后，本文详细地介绍了基于非负矩阵分解的语音增强方法的基本原理，包括训练阶段和增强阶段的具体过程，并进行了实验，此外，还利用一个基于非负矩阵分解的语音重构实验验证了语音基矩阵对语音频谱的建模能力。最后，本文总结了传统的基于非负矩阵分解的算法的不足，并对一些现有的基于非负矩阵分解的算法分别做了一个简单的概述，包括其创新点和优缺点，并对比分析了几种具有代表性的方法。本文从历史的角度展示了基于非负矩阵分解的语音增强方法的不断发展。      

基于约束非负矩阵分解的高光谱图像解混快速算法

约束非负矩阵分解是高光谱图像解混中常用的方法.该方法的求解通常采用投影梯度法,其收敛速度、求解精度和算法稳定性都有待提高.为此,本文针对较优的最小体积约束,提出一种基于约束非负矩阵分解的高光谱图像解混快速算法.首先优化原有的最小体积约束模型,然后设计了基于交替方向乘子法的非凸项约束非负矩阵分解算法,最后通过奇异值分解优化迭代步骤.模拟和实际数据实验结果验证了本文算法的有效性.     

基于贝叶斯阴阳机的2kb/s NMF-WI语音编码算法

 本文提出了一种改进型的基于非负矩阵分解(Nonnegative Matrix Factorization,NMF)的特征波形(Characteristic Waveform,CW)分解算法,一方面应用惩罚次胜者竞争学习算法(Rival Penalized Competitive Learning,RPCL)和贝叶斯阴阳机(Bayesian Ying-Yang,BYY)和谐学习算法,来计算NMF分解阶数,在没有明显降低语音质量的前提下,降低了编码器的复杂度;另一方面根据CW 的能量与编码矩阵的能量间的变化关系,提出了相位谱的混合自回归合成方法,提高了语音的自然度.最后,开发出一套改进型2kb/s NMF-WI低复杂度语音编码方法,采用基于K-L散度的NMF迭代算法和收敛速度更快的基矢量Mel刻度分带初始化方法,按照基音周期的统计分布将特征波形分为6类,在CW分解模块,复杂度下降了10MOPS,语音质量提高,与采用4bit散布矢量量化相位谱的2.16kb/s NMF-WI语音编码器的语音质量相当.     

基于增量式鲁棒非负矩阵分解的短文本在线聚类

对社会化媒体产生的大量短文本进行聚类分析具有重要的应用价值,但短文本往往具有噪音数据多、增长迅速且数据量大的特点,导致现有相关算法难于有效处理.提出一种基于增量式鲁棒非负矩阵分解的短文本在线聚类算法STOCIRNMF.STOCIRNMF基于非负矩阵分解构建短文本聚类模型,通过l_2,1范数设计模型的优化求解目标函数提高鲁棒性,同时应用增量式迭代更新规则实现短文本的在线聚类.在搜狐新闻标题和微博短文本数据集上进行相关实验,结果表明STOCIRNMF不仅比现有代表性算法具有更好的聚类性能,而且能够有效对微博话题进行在线检测.     

不完全非负矩阵分解的加速算法

非负矩阵分解(NMF)已成为数据分析与处理的一种日益流行的方法.当数据矩阵不完全时,可用加权非负矩阵分解(WNMF)来分解矩阵.但是在WNMF算法中,对于给定的搜索方向,步长的选取一般来说不是最优的.本文研究了不完全非负矩阵分解(INMF)问题,提出了加速算法(AINMF).首先,将INMF问题转化为交替地求解两个非负...     

基于二维非负矩阵分解的1kb/s WI语音编码算法

 本文针对波形内插（WI）语音编码模型和参数量化等技术进行了研究,并最终提出了一种基于二维非负矩阵分解的1kb/s波形内插（2DNMF-WI）语音编码算法. 文中采用二维非负矩阵分解（2D-NMF）方法来分解语音特征波形（CW）,该分解方法在行和列两个方向上同时压缩CW幅度谱矩阵的维数,使得CW幅度谱矩阵降维后得到的编码矩阵维数较小,易于量化. 此外,在甚低速率语音编码中,由于没有足够的比特数来描述编码参数,往往很难得到高质量的合成语音. 本算法采用两帧联合编码、帧间后向预测三级矢量量化、离散余弦变换（DCT）和分裂式矩阵量化等技术来降低编码速率和改善音质. 非正式主观听觉测试显示,1kb/s 2DNMF-WI编码器合成语音的质量稍差于2kb/s的NMF-WI语音编码算法.     

基于FFST和方向特性对比度的图像融合算法

为了克服红外可见光图像融合方法存在的不足,结合快速有限剪切波变换(Fast Finite Shearlet Transform,FFST)的平移不变性以及较高的方向敏感性,提出了一种基于快速有限剪切波变换域的自适应多方向图像融合新方法。首先,对严格配准后的图像进行快速有限剪切波变换分解,得到低频子带和高频子带系数;然后,对低频子带系数采用非负矩阵分解的一个约束稀疏算法,即在基本非负矩阵分解的优化函数中施加稀疏性约束,使分解更优,以此来提高重构后图像的清晰度;高频子带系数则采用联合方向特性的对比度进行选取,以得到丰富的细节信息。最后,利用快速有限剪切波逆变换得到重构后的图像。实验结果表明,融合后的图像充分结合了源图像的有用信息,整体轮廓清晰,在客观评价上也有一定的提高。     

松弛耦合非负矩阵分解的低分辨率人脸识别算法

针对实际监控场景中经常遇到的人脸图像分辨率较低的问题,本文提出了一种利用耦合非负矩阵分解并保持系数松弛的低分辨率人脸识别算法（Relaxed Coupled Nonnegative Matrix Factorization,后文简称RCNMF）。首先,对高低分辨率人脸图像进行非负矩阵矩阵分解（nonnegative matrix factorization,后文简称NMF）,在分解的同时保持组合系数近似一致,从而得到高低分辨率图像的基矩阵。然后,通过低分辨率图像的基矩阵提取训练和测试样本的特征。最后进行识别。实验结果验证了与其他几种基于耦合映射的低分辨率人脸识别方法相比,RCNMF算法的识别性能更好。同时通过实验验证了RCNMF算法的收敛性。      

全变差稀疏约束深度非负矩阵分解高光谱遥感影像解混方法

传统非负矩阵分解方法仅基于单层线性模型,现有的深度非负矩阵分解模型忽略了地物光谱的实际混合物理过程,仅从数学理论考虑深度分解。对此,文中从光谱混合的物理过程出发,综合非负矩阵分解和深度学习,将光谱混合过程进行反向建模,并充分考虑丰度的稀疏性和空间平滑性,构建了用于高光谱遥感影像解混的面向端元矩阵的全变差稀疏约束深度非负矩阵分解模型。通过模拟实验和真实实验,将文中所提方法与5种解混方法进行对比。结果表明,相较于面向丰度的深度非负矩阵分解算法,文中所提方法的平均光谱角距离和均方根误差均有所降低,取得了最佳解混结果。     

基于图流在线非负矩阵分解的社团检测

针对现有的在线社团检测方法大多仅从增量相关的节点和边出发,难以有效挖掘社团结构的动态变化特性问题,提出了一种基于图流在线非负矩阵分解的社团检测方法.首先将网络中持续到达的图数据按照流式数据进行存储和预处理,然后借鉴梯度下降思想,采用在线非负矩阵分解架构,根据不同时刻达到的图流序列,实时迭代更新社团归属矩阵,并通过有效的学习率和缓存策略设置,保证了图流处理的收敛性和合理性.实验结果表明,相比于已有在线社团检测方法,该方法具备更高的社团检测精度.     

基于角点检测和NMF的图像Hash算法

为了有效地实现图像Hash函数在图像认证检索中的应用,提出了结合Harris角点检测和非负矩阵分解(NMF)的图像Hash算法,首先提取图像中的角点,对角点周围图像块信息进行非负矩阵分解得到表征图像局部特征的系数矩阵,进一步量化编码产生图像Hash。实验结果表明,得到的图像Hash对视觉可接受的操作如图像缩放、高斯低通滤波和JPEG压缩具有良好的稳健性,同时能区分出对图像大幅度扰动或修改的操作。     

Robust Non‐negative Matrix Factorization with β‐Divergence for Speech Separation

This paper addresses the problem of unsupervised speech separation based on robust non‐negative matrix factorization (RNMF) with β‐divergence, when neither speech nor noise training data is available beforehand. We propose a robust version of non‐negative matrix factorization, inspired by the recently developed sparse and low‐rank decomposition, in which the data matrix is decomposed into the sum of a low‐rank matrix and a sparse matrix. Efficient multiplicative update rules to minimize the β‐divergence‐based cost function are derived. A convolutional extension of the proposed algorithm is also proposed, which considers the time dependency of the non‐negative noise bases. Experimental speech separation results show that the proposed convolutional RNMF successfully separates the repeating time‐varying spectral structures from the magnitude spectrum of the mixture, and does so without any prior training.     

Spectral factorization of two-variable para-Hermitian polynomial matrices

This paper presents an algorithm for the so-called spectral factorization of two-variable para-Hermitian polynomial matrices which are nonnegative definite on thej axis, arising in the synthesis of two-dimensional (2-D)passive multiports, Wiener filtering of 2-D vector signals, and 2-D control systems design. First, this problem is considered in the scalar case, that is, the spectral factorization of polynomials is treated, where the decomposition of a two-variable nonnegative definite real polynomial in a sum of squares of polynomials in one of the two variables having rational coefficients in the other variable plays an important role (cf. Section 4). Second, by using these results, the matrix case can be accomplished, where in a first step the problem is reduced to the factorization of anunimodular para-Hermitian polynomial matrix which is nonnegative definite forp=j , and in a second step this simplified problem is solved by using so-called elementary row and column operations which are based on the Euclidian division algorithm. The matrices considered may be regular or singular and no restrictions are made concerning the coefficients of their polynomial entries; they may be either real or complex.     

Robust Speech Hash Function

In this letter, we present a new speech hash function based on the non‐negative matrix factorization (NMF) of linear prediction coefficients (LPCs). First, linear prediction analysis is applied to the speech to obtain its LPCs, which represent the frequency shaping attributes of the vocal tract. Then, the NMF is performed on the LPCs to capture the speech's local feature, which is then used for hash vector generation. Experimental results demonstrate the effectiveness of the proposed hash function in terms of discrimination and robustness against various types of content preserving signal processing manipulations.     

Sparsity induced convex nonnegative matrix factorization algorithm with manifold regularization

To address problems that the effectiveness of feature learned from real noisy data by classical nonnegative matrix factorization method,a novel sparsity induced manifold regularized convex nonnegative matrix factorization algorithm (SGCNMF) was proposed.Based on manifold regularization,the L_2,1norm was introduced to the basis matrix of low dimensional subspace as sparse constraint.The multiplicative update rules were given and the convergence of the algorithm was analyzed.Clustering experiment was designed to verify the effectiveness of learned features within various of noisy environments.The empirical study based on K-means clustering shows that the sparse constraint reduces the representation of noisy features and the new method is better than the 8 similar algorithms with stronger robustness to a variable extent.     

Factorizations fornD polynomial matrices

In this paper, a constructive general matrix factorization scheme is developed for extracting a nontrivial factor from a givennD (n>2) polynomial matrix whose maximal order minors satisfy certain conditions. It is shown that three classes ofnD polynomial matrices admit this kind of general matrix factorization. It turns out that minor prime factorization and determinantal factorization are two interesting special cases of the proposal general factorization. As a consequence, the paper provides a partial solution to an open problem of minor prime factorization as well as to a recent conjecture on minor prime factorizability fornD polynomial matrices. Three illustrative examples are worked out in detail.     

基于四阶累积张量方法的欠定盲源信号分离

针对瞬时欠定盲源信号分离问题,提出一种四阶累积张量分解算法.首先构建观察信号四阶累积协方差,依据源信号具有相互独立且均值为零的性质,对累积协方差化简并扩展到张量域,得到四阶累积张量.采用分层交替最小二乘算法对四阶累积张量进行非负库克分解,求得非负库克模型的参数,同时获得非负混合矩阵并求其伪逆,最终估计出源信号.选用真实的语音信号和生物信号进行仿真实验,结果表明该方法提高了源信号和非负混合矩阵的估计性能.