（毕业设计论文）Gabor滤波的特征提取算法.doc

PAGE 1 Gabor滤波的特征提取算法 摘 要 人脸识别在计算机视觉和模式识别领域是最具挑战性的研究课题，在近20 年内提出了许多经典的算法。如Meng Yang 等人使用Gabor 小波变换提取局部特征用来稀疏表示（GSRC）[4], 使得人脸识别率明显提高。但这种方法直接对Gabor 变换后的能量子带进行特征融合，不仅使得能量分布比较分散，而且使得结构信息分散，不利于对局部特征的提取。本文针对以上问题，提出了基于Gabor特征的稀疏表示人脸识别算法，其思路是首先对每幅人脸图像采用Gabor小波变换进行特征提取，并将计算得到的Gabor特征向量作为人脸图像的初始特征，从而减弱光照和噪声对图像的影响；然后通过稀疏表示对Gabor特征向量进行优化，构成增强的Gabor 特征。从而建立一个稀疏且能准确反映人脸图像的特征向量用于人脸区分，最后将该特征应用于稀疏表示人脸识别。 论文首先通过对人脸傅里叶变换和Gabor变换的实验比较，证明了Gabor变换在提取人脸特征方面具有很大的优越性。其次，针对Gabor和LBP提取的人脸特征向量维数过高问题，分别采用了LPP和PCA降维算法来进行降维，来改善Gabor局部特征的提取能量分布，结构信息分散的缺点。最后，将提取出来的人脸特征向量进行验证，并对两种不同降维方法做比较。并在orl和yale两个人脸图库做仿真实验。实验表明，采用用LPP降维得到的相似度要远高于使用PCA降维结果。 关键词：人脸识别；Gabor小波变换；稀疏表示；LBP Said the sparse face recognition based on Gabor features Abstract Face recognition in the field of computer vision and pattern recognition is the most challenging research topic in the recent 20 years, put forward many classical algorithms. Such as Meng Yang and Gabor wavelet transform is used to extract local features for sparse representation (GSRC) [4], the face recognition rate is higher. But this method directly on the Gabor transformed energy subband feature fusion, not only makes the energy distribution of scattered, but also makes the structure information is scattered, not conducive to the extraction of local features. In this paper, in view of the above problems, said face recognition algorithm is proposed based on sparse Gabor features, the idea is to first use of Gabor wavelet transform for each face image feature extraction, and the Gabor feature vector is calculated as the initial feature of face images, thereby reducing the effects of weak light and noise on the image; and then through the sparse said to optimize Gabor feature vector, Gabor feature enhancement. To build a sparse and can accurately reflect the feature vector of face image for face discrimination, the features used in sparse representation for face recognition. Firstly, through the comparison of the face of Fourier transform and Gabor trans