人脸识别问题的特点包括样本的特征维数高和每个类别所包含的样本较少。设计有效的特征提取方法是解决人脸识别问题的关键要素之一。提出了在采用降采样获得特征的同时利用新的降采样方法多次对原图片进行降采样, 生成多幅训练样本, 进而缓解人脸识别中的小样本问题。实验结果证明所提出的方法能有效地提高分类器的精度。

The characteristic that samples are high dimensional and small is typical for face recognition (FR). It is a key point to develop effective feature extraction approach for FR problem. A method to produce multiple copies of a face image is proposed while doing downsampling on original image. The experiment indicates that the proposed method can improve the recognition accuracy of classifiers.