Abstract: For face recognition with single training sample per person，the conventional face recognition methods which work with many training samples don’t function well.Especially，a number of methods based on Fisher linear discriminant criterion can’t work because the within-class scatter matrix is a matrix with all elements being zero.To overcome the above problem，a new sample augment method is proposed in this paper.It slices the image at eight different planes（bit-planes）.It augments new training samples by combining the bit-planes images.Two-dimensional maximum scatter-difference discriminant analysis is performed on the new training images obtained.The experimental results on ORL face database show that the proposed method is effective and promising in face recognition with single training sample per person.

摘要： 在进行单训练样本人脸识别时，基于每人多个训练样本的传统人脸识别算法效果通常不太理想。尤其是基于Fisher线性鉴别准则的一些方法，由于类内散布矩阵为零矩阵，根本无法进行识别。针对以上问题进行了分析研究，提出了一种新的样本扩充方法，即：采用位平面图像分解法，将每幅样本图像分解为8幅，进而通过各种合成策略构造多幅样本图像。使用一种更加稳定的二维最大散度差线性鉴别分析方法（2DMSLDA）对上面获得的新样本图像进行特征抽取。在ORL国际标准人脸库上进行的实验表明了所提算法的可行性和有效性。