摘要
为了克服光照强度和方向等变化带来的误差干扰,提出了一种基于标签一致K-SVD(LC-KSVD)字典学习的人脸识别方法。首先对训练样本进行直方图均衡和小波去噪相结合的图像预处理;然后利用主成分分析对图像进行降维及特征提取,构建初始字典;接下来在字典学习过程中引入约束项,得到有判别能力的新字典;进而,计算测试样本在新字典下的稀疏系数,并进行类关联重构;最后根据重构误差,完成分类识别。实验结果验证了改进算法的有效性。
In order to overcome the error interference caused by changes in intensity and direction of illumination,a face recognition method based on label consistent( LC)-KSVD dictionary learning is proposed.Firstly,the image is preprocessed by histogram equalization and wavelet denoising. Then,the PCA reduce dimensionality and extract feature on the image to construct an initial dictionary. Next,we introduce constraints in the process of dictionary learning. The new dictionary with discriminating ability is obtained. Then,the sparse coefficients of test samples under the new dictionary are calculated and the class association reconstruction is carried out. Finally,the recognition is completed according to the reconstruction error. The experimental results demonstrate the effectiveness of the improved algorithm.
作者
严春满
张迪
郝有菲
陈佳辉
胡志斌
YAN Chunman;ZHANG Di;HAO Youfei;CHEN Jiahui;HU Zhibin(School of Physics and Electronic Engineering,Northwest Normal University,Lanzhou 730070,China)
出处
《传感器与微系统》
CSCD
2020年第11期44-46,共3页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61741119)
甘肃省自然科学基金资助项目(17JR5RA074,17JR5RA078)。
关键词
人脸识别
直方图均衡
小波去噪
标签一致K-SVD算法
face recognition
histogram equalization
wavelet denoising
label consistent(LC)-KSVD algorithm
