Abstract: A novel method based on three-layer virtual image generation for face recognition with single sample is presented. Singular Value Decomposition(SVD)-perturbation is used to highlight the facial features, and the method of geometric transformation is used to enhance the changes of attitude and scale. The number of sample is increased. The method based on spatial distribution is used to improve the distribution of samples, so that the virtual sample distribution is more close to the real-world distribution. Experimental results carried on ORL face database validate the efficiency of the method.

Key words: Principal Component Analysis(PCA), Singular Value Decomposition(SVD)-perturbation, virtual image generation, single sample, face recognition

摘要： 针对人脸识别中的单训练样本情况下识别率较低的问题，提出一种三层虚拟图像生成方法。采用奇异值扰动方法突出人脸特征，通过几何变换方法增强姿态、尺度变化和样本数量，基于空间分布的方法改善样本分布。在ORL人脸库上的实验结果表明，该方法能有效地对单样本问题中的训练样本进行预处理。

关键词: 主成分分析, 奇异值扰动, 虚拟图像生成, 单样本, 人脸识别

CLC Number: 

• TP18