基于变化稀疏表示的单样本人脸识别

One Sample per Person Face Recognition via Variation Sparse Representation

提出一种基于变化稀疏表示的单样本人脸识别算法,将测试图像相对于某一标准图像的人脸变化表示为类内及类间变化的线性组合,通过求解最小L1范数得到线性组合的稀疏表示系数。识别时,对应于类间变化字典中最大稀疏表示系数的变化样本给出了测试图像的身份信息。算法在公共测试库Extended Yale Face Database B上的实验结果证明,该算法在得到优于或相近于ESRC及ELRC识别率的同时,运算时间少于ESRC及ELRC算法。

In this paper we propose a new model to describe the mixed variation between the test image and the natural image. The mixed variation is described as a linear combination of intra-class and inter-class variations. The sparse coefficients of combination are obtained by solving L1-norm minimization problem. Identity of query image is implied by the atom corresponding to the biggest coefficient in the inter-class variation dictionary. Experiments on public face database （Extended Yale Face Database B） are conducted to prove the validity of the proposed algorithm, and the experimental results show that our algorithm demonstrates promising recognition performance and costs less time.