《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (4): 1013-1020.DOI: 10.11772/j.issn.1001-9081.2022030329

• 人工智能 • 上一篇    

基于近邻监督局部不变鲁棒主成分分析的特征提取模型

葛孟婷1, 万鸣华1,2()   

  1. 1.南京审计大学 计算机学院(智能审计学院), 南京 211815
    2.江苏省社会安全图像与视频理解重点实验室(南京理工大学), 南京 210094
  • 收稿日期:2022-03-21 修回日期:2022-06-18 接受日期:2022-06-27 发布日期:2023-04-11 出版日期:2023-04-10
  • 通讯作者: 万鸣华
  • 作者简介:葛孟婷(1998—),女,江苏连云港人,硕士研究生,CCF会员,主要研究方向:模式识别、特征提取、人脸识别、低秩学习;
  • 基金资助:
    国家自然科学基金资助项目(61876213);江苏省自然科学基金资助项目(BK20201397);江苏省社会安全图像与视频理解重点实验室项目(J2021?4);江苏省高校未来网络科研基金资助项目(SRFP?2021?YB?25);2021年江苏省研究生科研与实践创新计划项目(SJCX21_0885)

Feature extraction model based on neighbor supervised locally invariant robust principal component analysis

Mengting GE1, Minghua WAN1,2()   

  1. 1.School of Computer Science (School of Intelligence Audit),Nanjing Audit University,Nanjing Jiangsu 211815,China
    2.Jiangsu Key Laboratory of Image and Video Understanding for Social Safety,Nanjing University of Science and Technology,Nanjing Jiangsu 210094,China
  • Received:2022-03-21 Revised:2022-06-18 Accepted:2022-06-27 Online:2023-04-11 Published:2023-04-10
  • Contact: Minghua WAN
  • About author:GE Mengting, born in 1998, M. S. candidate. Her research interests include pattern recognition, feature extraction, face recognition, low-rank learning.
  • Supported by:
    National Natural Science Foundation of China(61876213);Natural Science Foundation of Jiangsu Province(BK20201397);Project of Jiangsu Key Laboratory of Image and Video Understanding for Siocial Safety(J2021-4);Qing Lan Project of Jiangsu University, Future Network Scientific Research Fund Project of Jiangsu Colleges and Universities(SRFP-2021-YB-25);2021 Postgraduate Research and Practice Innovation Program of Jiangsu Province(SJCX21_0885)

摘要:

针对无监督的局部不变鲁棒主成分分析(LIRPCA)算法未考虑样本间的类别关系的问题,提出了一种基于近邻监督局部不变鲁棒主成分分析(NSLIRPCA)的特征提取模型。所提模型考虑了样本间的类别信息,并以此构建关系矩阵。对所提模型进行公式求解和公式的收敛性证明,并将所提模型应用于各种遮挡数据集。实验结果表明,在ORL、Yale、COIL-Processed和PolyU数据集上,与主成分分析(PCA)算法、基于L1范数的主成分分析(PCA-L1)算法、非负矩阵分解(NMF)算法、局部保持投影(LPP)算法和LIRPCA算法相比,所提模型在原始图像数据集上的识别率分别最高提升了8.80%、7.76%、20.37%、4.72%和4.61%,在遮挡图像数据集上的识别率分别最高提升了30.79%、30.73%、36.02%、19.65%和17.31%。可见,所提模型提高了算法的识别性能,降低了模型复杂度,明显优于对比算法。

关键词: 特征提取, 子空间学习, 无监督学习, 鲁棒性, 图像识别

Abstract:

Focused on the issue that the category relationship between samples is not considered in the unsupervised Locally Invariant Robust Principal Component Analysis (LIRPCA) algorithm, a feature extraction model based on Neighbor Supervised LIRPCA (NSLIRPCA) was proposed. The category information between samples was considered by the proposed model, and a relationship matrix was constructed based on this information. The formulas of the model were solved and the convergences of the formulas were proved. At the same time, the proposed model was applied to various occlusion datasets. Experimental results show that compared with Principal Component Analysis (PCA), PCA based on L1-norm (PCA-L1), Non-negative Matrix Factorization (NMF), Locality Preserving Projection (LPP) and LIRPCA algorithms on ORL, Yale, COIL-Processed and PolyU datasets, the proposed model has the recognition rate improved by 8.80%, 7.76%, 20.37%, 4.72% and 4.61% at most respectively on the original image datasets, and the recognition rate improved by 30.79%, 30.73%, 36.02%, 19.65% and 17.31% at most respectively on the occluded image datasets. It can be seen that with the proposed model, the recognition performance of the algorithm is improved, and the complexity of the model is reduced, verifying that the model is obviously better than the comparison algorithms.

Key words: feature extraction, subspace learning, unsupervised learning, robustness, image recognition

中图分类号: