基于小波函数的二维整体经验模态分解研究

发布时间：2018-05-01 11:34

本文选题：经验模态分解 + 表情识别　；参考：《长春工业大学》2017年硕士论文

【摘要】：现今,黄锷和他的团队所提出的希尔伯特黄变换(HHT)成为了各大领域的研究热点,HHT方法在1998年提出以来,经过国内外学者们的深入研究,取得了一系列的研究成效。经验模态分解EMD方法是希尔伯特黄变换中最重要的且具有创新性的数据处理算法。自适应性是EMD算法最主要的特性之一,它主要是将非线性、非平稳数据序列或信号的进行平稳化。由于EMD方法的提出以及广大学者的广泛关注,该算法已经快速有效的应用在各种的工程领域,并取得一定的研究成果。自适应性是EMD算法最主要的特性之一,它主要是将非线性、非平稳数据序列或信号的进行平稳化。EMD可把复杂的数据分解成有限的本征模函数(IMF)和一个趋势项,从而使得瞬时频率这一概念具有了实际的物理意义。但是EMD算法存在着模态混叠的问题。为了克服了这一缺点,提出了一种改进算法——整体经验模态分解(EEMD)方法,它是针对EMD方法的缺点而提出的一种噪声辅助数据分析方法。为了将该算法应用在图像处理方面,本文提出了一种基于小波函数的二维整体经验模态分解的方法,主要采用小波函数为二维EEMD算法中的基函数,利用最小二乘法原理实现自适应数据拟合。本文以人脸表情图像为研究对象,证明所提出的算法是可行的并且有效的。本文针以人脸面部表情作为研究对象进行研究分析,所研究的人脸表情来源于日本JAFFE人连表情数据库,其中的213个表情来自10个不同的女性。本文利用图像的尺度归一化和直方图均衡化对表情图像进行预处理,对于处理的结果,首先进行Radon变换,再利用一维经验模态分解对人脸图像作进一步处理;其次,将预处理的结果再进行二维经验模态分解;最后,分别对一维和二维经验模态分解的结果采用支持向量机的方法,将表情特征属性数据进行训练分类,对比其分类结果并进行分析,找到更加有效的人脸面部表情都识别的方法。其分析结果对未来的图像处理方面,特别是人脸表情识别将会有重要的参考价值。
[Abstract]:Nowadays, HHT proposed by Huang E and his team has become a research hotspot in various fields. Since 1998, the HHT method has been studied deeply by scholars at home and abroad, and a series of research results have been achieved. Empirical mode decomposition (EMD) EMD method is the most important and innovative data processing algorithm in Hilbert-Huang transform. Self-adaptability is one of the most important features of EMD algorithm. It mainly stabilizes nonlinear, non-stationary data sequences or signals. Due to the development of EMD method and the wide attention of many scholars, the algorithm has been applied in various engineering fields quickly and effectively, and some research results have been obtained. Self-adaptability is one of the most important features of EMD algorithm. It can decompose complex data into finite eigenmode function and a trend term by stabilizing nonlinear, non-stationary data sequences or signals. Thus, the concept of instantaneous frequency has practical physical significance. But the EMD algorithm has the problem of modal aliasing. In order to overcome this shortcoming, an improved algorithm, Global empirical Mode decomposition (EMD) method, is proposed, which is a noise-aided data analysis method aiming at the shortcomings of EMD method. In order to apply this algorithm to image processing, this paper presents a method of two-dimensional global empirical mode decomposition based on wavelet function. The wavelet function is used as the basis function of two-dimensional EEMD algorithm. The principle of least square method is used to realize adaptive data fitting. This paper takes facial expression image as the research object, and proves that the proposed algorithm is feasible and effective. In this paper, facial expressions are studied and analyzed. The facial expressions are derived from the JAFFE Human expression Database in Japan, and 213 of them come from 10 different women. In this paper, the scale normalization and histogram equalization of the image are used to preprocess the facial expression image. First, the Radon transform is carried out, then the one-dimensional empirical mode decomposition is used to further process the face image. Finally, the results of one-dimensional and two-dimensional empirical mode decomposition are analyzed by support vector machine (SVM), and the facial expression attribute data are trained and classified. The classification results are compared and analyzed to find a more effective method for facial expression recognition. The analysis results will have important reference value for future image processing, especially for facial expression recognition.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】