中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
自适应信号与图像分解及其应用

文献类型:学位论文

作者胡晰远
学位类别工学博士
答辩日期2011-05-31
授予单位中国科学院研究生院
授予地点中国科学院自动化研究所
导师彭思龙
关键词经验模态分解 模式混叠 基于算子的方法 积分算子 图像分解 图像压缩 Empirical mode decomposition mode mixing operator-based approach integral operators image decomposition image compression
其他题名Adaptive Signal and Image Separation and its Applications
学位专业模式识别与智能系统
中文摘要近年来,单通道信号分解因其在众多领域中的广泛应用而倍受关注。在绝大部分的单通道信号分解算法中,都是将输入信号建模为多个基本信号之和的形式。不同的方法由于对基本成分的定义不同而导致了不同的分解结果。在所有这些方法中,经验模态分解(Empirical Mode Decomposition,EMD)和基于算子(Operator-Based)的信号分解算法由于它们的完全由数据驱动的自适应性而受到我们的重点关注。 对于EMD算法而言,停止条件、包络算法及模式混叠问题是影响它性能的三个关键因素。在本论文中,我们通过多分量调幅调频(Amplitude Modulated and Frequency Modulated,AM-FM)信号模型来深入分析EMD算法中的包络技术和模式混叠问题。在指出了以往研究中有关包络的一个不合适的假设后,我们给出有关包络的一个新的必要条件。在此基础上,我们提出了一种新的计算包络的数值方法,实验证明该包络算法是有效的。然后,我们提出了两种方法用来解决EMD算法中存在的模式混叠问题。第一种方法在基于每个本征模 式函数(Intrinsic Mode Function,IMF)之间都是局部正交的假设下,提出了一种背投影算法。尽管这种方法能解决绝大部分情况下的模式混叠问题,但由于它采用了诸如Gabor变换这类的时频分析工具,使得它受到了算法复杂度和测不准原理的限制。因此,我们又提出了第二种完全在时间域处理的算法,这种算法能够有效地解决在多分量AM-FM信号分解中存在的模式混叠问题。 基于算子的自适应信号分解算法是一种根据所定义的算子,通过优化的方式来对信号进行分解的算法,它的优点在于,可以根据目标信号的特点设计出与之对应的算子。本论文中,一方面,我们提出了一类新的微分算子来实现多分量AM-FM信号的分解与解调,它是对零空间追踪(Null Space Pursuit,NSP)算法中所采用的微分算子形式的改进。另一方面,通过第二类Fredholm积分方程,我们定义了一种新的积分算子并给出了它的一般形式。微分算子能够刻画信号的振荡性,积分算子则能够很好的保证信号的对称性,所以通过微分算子和积分算子,我们能更全面的刻画信号的不同特性。接着,我们还详细分析了所提出的积分算子的相关性质,并将其与IMF定义中的第二个条件及短时傅里叶变换之间进行了比较;还将积分算子引入到NSP算法中,对多分量信号的分解进行了实验比较。实验表明,我们新定义的积分算子具有很好的有效性和鲁棒性。 最后,我们将一维信号中的算子方法推广到了二维的图像分解中。通过对分解出来的图像成分加入不同的先验约束,提出了两种基于算子的图像分解算法,即图像的卡通+纹理分解算法和快速的图像线性分解算法。之后,我们将所提出的快速的图像分解算法应用到图像压缩领域中,构建了一种全新的静态图像的多分量预测编码框架。在该框架中,我们首先将重构图像分解为若干个子成分,然后采用不同的预测模式分别对每个子成分单独进行预测,再对它们的残差统一进行编码。采用多分量预测算法能够很大程度地提高对图像中像素值预测的准确性。通过使用H.264/AVC中的残差编码算法,我们将所提出的多分量预测编码算法与目前最...
英文摘要Single channel signal separation has attracted a great deal of attention in recent years since it has affected many applications. Many approaches have been proposed to separate a single channel signal into a mixture of several additive coherent subcomponents. The methods used to separate signals vary because the different definitions of subcomponents are used. Among those approaches, the Empirical Mode Decomposition (EMD) and operator-based method are of particular interest to us, since both of them are self adaptive and fully data driven approaches. To improve the EMD algorithm, the stop criterion, envelope technique, and mode-mixing problem are the three most important topics that need to be addressed. In this thesis, we study the envelope technique and the mode-mixing problem caused by separating multicomponent AM-FM signals with the EMD algorithm. We present a new necessary condition on the envelope that questions the current assumption that the envelope passes through the extrema points of an intrinsic mode function (IMF). Then, we present two solutions to resolving the mode-mixing problem which can be viewed as a generalization of Deering and Kaiser’s method. The first approach uses a back projection strategy, which is based on the assumption that each IMF should be locally orthogonal to the others. Although it is very effective in resolving the mode mixing problem for a broad range of signals, it relies on some time-frequency analysis techniques such as Gabor transform. And then, the second approach, which is a totally time domain method, has been proposed for solving the mode mixing phenomenon that occurs when multicomponent AM-FM signals are separated. The operator-based signal separation approach, which can be formulated as an optimization problem, uses an adaptive operator to separate a signal into additive subcomponents. The charming feature of using the approach to solve the signal separation problem is that the design of the operator can be customized to the target signal. In this thesis, on the one hand, we propose a new kind of differential operator to separate a multicomponent AM-FM signal; and then, we use the estimated operators to calculate each sub-component’s envelope and instantaneous frequency. This proposed differential operator can be viewed as the improvement of the Null Space Pursuit algorithm. On the other hand, considering different types of operators can characterize different properties of a signal, we define a new k...
语种中文
其他标识符200718014628036
源URL[http://ir.ia.ac.cn/handle/173211/6379]  
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
胡晰远. 自适应信号与图像分解及其应用[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2011.

入库方式: OAI收割

来源:自动化研究所

浏览0
下载0
收藏0
其他版本

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。