Abstract: This paper presents a method for face retrieval in video stream based on SVD and improved PCA. The PCA is improved through local mean and standard deviation in order to overcome the effects of illumination. The AdaBoost is used to detect human faces in image and video. The training samples are increased by Singular Value Decomposition（SVD）. On the basis of the original and new samples, the algebra features are extracted by using improved Principal Component Analysis（PCA）. The features are compared through nearest neighbor classifier and the retrieval results are displayed to users. Experimental results show the method performs well in simple background videos.

Key words: face detection, single sample face recognition, Singular Value Decomposition（SVD）, Principal Component Analysis（PCA）, video-based face retrieval

摘要： 针对视频中人脸检索问题，提出一种基于奇异值分解和改进PCA相结合的视频中单样本人脸检索方法，其中通过融合局部均值和标准差的图像增强处理来实现PCA算法的改进，从而克服光照对目标的影响。通过AdaBoost人脸检测算法对人脸图像和视频进行人脸检测；通过奇异值分解增加训练样本，在原样本和新样本的基础上采用改进的PCA人脸识别算法提取待检测人脸和视频中的人脸代数特征；采用最近邻分类器进行特征匹配，判断视频中检测出的人脸是否为要检索的目标人脸。实验结果表明，该方法在简单背景的视频环境下可以较准确地检索出目标人脸。

关键词: 人脸检测, 单样本人脸识别, 奇异值分解, 主分量分析, 基于视频的人脸检索