摘要
双密度双树复小波对图像细节信息描述更加清晰,使用双密度双树复小波对人脸图像进行信息提取,可使图像中的信息更好的保留下来。针对Gabor小波进行特征提取时的频域重叠以及频域遗漏和双树复小波提取的方向信息确失,提出利用两种小波对人脸图像提取的特征信息进行组合,得到人脸图像的特征信息。再针对提取到的特征信息维数过多不利于信息处理,提出了利用LLTSA算法对经过小波变换得到的特征信息进行降维。在降维的过程中先利用改进的局部切空间排列算法进行降维,然后使用三阶近邻算法进行分类。实验结果表明:使用该算法在ORL及YALE人脸库进行的人脸识别较传统的识别算法的识别结果更好,识别速率更高。
Dual-density tree complex wavelet will obtain more clearer image details, so many researchers use it for face image information extraction which can be better preserved. However, due to the restriction of its direction and the overlap and omissions for frequency-domain information which is obtained from Gabor wavelet feature extraction, we combine the Dual-density tree complex wavelet and Gabor wavelet to extract the feature information for the face images. For the extracted feature information has a large number of dimension which is not conducive to information processing, so we try to use lltsa （Linear Local Tangent Space Alignment） to reduce the dimension information that is obtained through the wavelet transform. In the process of dimension reduction ,we first use the improved local tangent space alignment algorithm to reduce the dimension, and then use the third-order nearest neighbor algorithm for classification. Experimental results show that：this face recognition algorithm in ORL and YALE face database recognition have better results and higher recognition rate than traditional recognition algorithm.
出处
《电子设计工程》
2015年第17期179-181,共3页
Electronic Design Engineering
基金
河北工业大学博士科研项目(2009032012)
关键词
人脸识别
特征提取
双密度双数复小波
降维
局部切空间排列算法
face recognition
feature extraction
dual density double tree wavelet
dimensionality reduction
LLTSA
