摘要
模块主成分分析是人脸重建中一种重要的子空间学习方法,鲁棒性不足是传统的基于L2范数的模块主成分分析(BPCA-L2)的主要问题。为此,提出了一种新的基于L1范数的模块主成分分析(BPCA-L1)方法。该方法使用了对奇异值不太敏感的L1范数。基于L1范数的模块主成分分析方法简单并易于实现,在一些人脸数据集上的重建实验验证了其有效性。
The block principal component analysis is an important subspace learning method in face reconstruction. Lacking robustness is a main problem of the traditional L2-norm (L2-BPCA). In this paper, a method of block principal component analysis (BPCA) based on a new L1-norm is introduced. L1-norm is used, which is less sensitive to abnormal values. The proposed block principal component analysis based on L1-norm is simple and easy to be implemented. Experimental reconstruction on several face databases are conductive to verifying the validity of L1-BPCA.
出处
《计算机时代》
2015年第2期24-25,28,共3页
Computer Era
基金
绍兴文理学院大学生科研基金项目(2013)