二维特征提取方法分析-信号与信息处理专业论文.docxVIP

摘 要 人脸识别技术是生物特征识别技术的一个重要分支，特征提取是人脸识别中的一个重 要方面，特征提取的好坏直接关系到人脸识别的最终效果。传统的特征提取主要关注于一 维特征提取的研究，本文在已有二维特征提取方法的基础上，重点研究了二维特征融合算 法和二维表示的稀疏保持投影算法，主要工作如下： (1) 在二维局部保持投影算法和二维线性鉴别分析算法的基础上，利用典型相关分析的 相关理论，本文将两种二维算法提出的特征集进行融合，提出了新的融合方法即 2DCLPP 算法，并在此基础上做了一定的改进。 (2) 在稀疏保持投影的基础上，将其扩展到二维特征提取领域，本文提出了二维表示的 稀疏保持投影算法即 2DSPP 算法，该算法可以有效地提取特征和解决协方差矩阵奇异问题； 此外，在 2DSPP 算法中，首先必须对图片进行二维稀疏表示，本文证明了二维稀疏表示系 数和一维稀疏表示系数的一致性，从而将问题转化为求较为容易的一维稀疏表示系数。 最后，本文分别在 ORL 人脸库，AR 人脸库及 Extended Yale 人脸库上验证了 2DCLPP 算法和 2DSPP 算法的有效性。实验表明，2DCLPP 是一种有效的特征融合办法，提取到的 特征可以有效地用于分类；2DSPP 是一种有效的二维特征提取方法，可以取得较好的识别 效果。 关键词： 二维特征提取；特征融合；典型相关分析；稀疏表示；二维稀疏保持投影； ABSTRACT Face recognition is an important branch of biometrics technology, feature extraction is an important aspect of face recognition, a good or bad method of feature extraction is directly related to the final result of recognition effect. The main concern of traditional feature extraction is one-dimensional research, Based on the analysis of the existing two-dimensional feature extraction methods, the paper focuses on two-dimensional feature fusion algorithm and two-dimensional sparse preserving projection algorithm. The main work is as follows: Based on two-dimensional locality preserving projection algorithm and two-dimensional linear discriminant analysis, This paper fuse two feature sets of two algorithm and the paper proposed a new fusion algorithm called 2DCLPP under the theory of canonical correlation analysis. In addtion, the paper proposed some suggestions for improvement. Based on locality preserving projection algorithm, the paper proposed two-dimensional locality preserving projection algorithm (2DSPP). The algorithm can effectively extract features and resolve the problem of singular covariance matrix. In addtion, the step of two-dimensional sparse representation is necessary in 2DSPP. It proved that the sparse coefficient of two-dimensional sparse representation is the same as one dimension in this paper. So it is easier to find the sparse coefficient f