基于压缩感知的单样本人脸识别

Compressive sensing-based face recognition for single sample

提出一种基于压缩感知的单样本人脸识别方法,通过局部邻域嵌入非线性降维和稀疏系数的方法产生冗余样本,则新样本包含了多种姿态和多种表情。将所有的新样本作为训练样本,运用改进后的稀疏表征分类算法进行人脸图像的识别。在单样本情况下,基于ORL人脸库和FERET人脸库的实验证明,该方法比原稀疏表征方法在识别率上分别提高了15.53%和7.67%。与RSRC、SSRC、DMMA、I-DMMA等方法相比,该方法同样具有良好的识别性能。

This paper proposes a kind of face recognition method with one training image per person, which is based on compressed sensing. There are two methods nonlinear dimensionality reduction by locally linear embedding and sparse coefficients, by witch redundant samples can generate. These new samples with multi-expressive and multi-gesture can be treated as training samples. Finally, the improved SRC algorithm can be applied to face recognition. Experiments on the well-known ORL face database and FERET face database show that the proposed method is respectively about 15.53% and 7.67%, more accurate than original SRC method in the context of single sample face recognition problem. In addition, extensive experimentation reported in this paper suggests that the proposed method achieves higher recognition rate than RSRC, SSRC, DMMA, and I-DMMA.