多矩阵低秩分解的鲁棒特征提取

doi:10.16337/j.1004-9037.2021.03.006

首页 > 按月查看>2021年第3月 >477-488. DOI:10.16337/j.1004-9037.2021.03.006

多矩阵低秩分解的鲁棒特征提取
DOI:
                        10.16337/j.1004-9037.2021.03.006
                    
作者:
                        
                        
                    
作者单位:1.北京建筑大学理学院，北京 100044;2.北京建筑大学北京未来城市设计高精尖创新中心，北京 100044
作者简介:
通讯作者:
基金项目:国家自然科学基金（62072024，61971290）资助项目；北京建筑大学青年教师科研能力提升计划（X21024）资助项目；北京建筑大学“建大英才”（JDYC2017026）资助项目；北京建筑大学北京未来城市设计高精尖创新中心（UDC2019033324）资助项目；北京建筑大学基本科研业务费（X20084）资助项目。

Robust Feature Extraction Based on Multi-matrix Low-Rank Decomposition

Author:

Affiliation:

1.School of Science, Beijing University of Civil Engineering and Architecture, Beijing 100044, China;2.Beijing Advanced Innovation Center for Future Urban Design,Beijing University of Civil Engineering and Architecture, Beijing 100044, China

Fund Project:

摘要

图/表

访问统计

参考文献

相似文献

引证文献

资源附件

摘要:

由于传统的人脸识别算法效果容易受制于光照、表情、遮挡以及稀疏大噪声等外界因素的影响，如何有效提取数据特征、进一步提升算法的鲁棒性，是传统人脸识别方法发展的关键所在。本文将多矩阵低秩分解应用在人脸特征提取中，充分利用多张人脸之间的结构相似性，探索人脸图像集的低秩子空间，进而结合低秩矩阵恢复模型来提取测试样本的低秩特征。最后，利用主成分分析（Principal component analysis， PCA）算法对所提取的特征矩阵进行进一步降维，并运用稀疏表示方法分类。实验结果表明，当样本中存在一定的椒盐噪声时，本文算法在AR、Yale和CMU_PIE人脸库上均具有较好的识别精度，验证了本文算法对椒盐噪声的鲁棒性。

Abstract:

Traditional face recognition algorithms are easily affected by lighting， expressions， occlusion， and sparse noise. How to effectively extract data features for traditional face recognition algorithms is one of the most important parts. This paper applies the multi-matrix low-rank decomposition to facial feature extraction， which makes full use of the structural similarity of face datasets and explores the low-rank subspace of the facial images collection， then combines the low-rank matrix recovery model to extract the key features of the test sample. Finally， the principal component analysis （PCA） algorithm is used to reduce the dimensionality of data， and the sparse representation is utilized for classification. The results show that the algorithm in this paper has good recognition accuracy on AR， Yale and CMU_PIE face datasets when samples contain salt and pepper noise， which verifies the robustness of the algorithm to salt and pepper noise.

表 1 AR库特征提取+PCA降维至150的实验数据Table 1 Experimental data of feature extraction of AR dataset with PCA reduced dimension to 150

表 2 Yale库特征提取+PCA降维至40的实验数据Table 2 Experimental data of feature extraction of Yale dataset with PCA reduced dimension to 40

表 4 Yale库样本添加椒盐噪声浓度为30%的实验数据Table 4 Experimental data of Yale sample added 30% salt and pepper noise

图1 AR 库下采样的样本（第1行为训练组，第2行为测试组）Fig.1 Images in AR dataset (The first line is training group and the second line is test group)

图2 AR 库样本依次加入 5%、10%、15%、20%、25%及 30% 的椒盐噪声Fig.2 AR samples with 5%,10%,15%,20%,25% and 30% pepper and salt noise

图3 AR库特征提取+PCA降维实验对比图Fig.3 Feature comparison of AR dataset feature extraction with PCA dimension reduction experiment

图4 Yale 库下采样的样本依次加入 10%、20%、30%、40%、50%及 60%的椒盐噪声Fig.4 Yale samples with 10%,20%,30%,40%,50% and 60% pepper and salt noise

图5 Yale库特征提取+PCA降维实验对比图Fig.5 Feature comparison of Yale dataset feature extraction with PCA dimension reduction experiment

图6 CMU_PIE人脸库样本Fig.6 CMU_PIE face library samples

图7 CMU_PIE库样本依次加入5%、10%、15%、20%、25%、30%的椒盐噪声Fig.7 CMU_PIE library added with 5%,10%,15%,20%,25% and 30% pepper and salt noise

图8 CMU_PIE 库特征提取+PCA 降维实验对比图Fig.8 Feature comparison of CMU_PIE library feature extraction with PCA dimension reduction experiment

表 5 CMU_PIE库特征提取+PCA降维至160的实验数据Table 5 Experimental data of CMU_PIE feature extraction with PCA reduced dimension to 160

表 3 Yale库特征提取+PCA降维至60的实验数据Table 3 Experimental data of feature extraction of Yale dataset with PCA reduced dimension to 60

参考文献

相似文献

引证文献

引用本文

米雪荣,王恒友,何强,张长伦.多矩阵低秩分解的鲁棒特征提取[J].数据采集与处理,2021,36(3):477-488

复制

文章指标

点击次数:
下载次数:

历史

收稿日期:2020-07-01
最后修改日期:2020-10-09
录用日期:
在线发布日期: 2021-05-25

引用本文

分享

文章指标

历史