摘要
提出一种视频图像中人脸识别的方法。首先将视频分解成单帧图像;然后利用基于肤色的方法计算人脸占图像的比例,进而确定关键帧位置;最后运用SIFT算法进行图像匹配,按照样本训练的阈值输出匹配结果。算法研究表明:该算法不需要大量视频训练并且保证了人脸识别系统的准确性,关键帧提取简单,相对传统的关键帧提取算法缩短了处理时间,为将人脸识别系统实时化提供了支持。其中视频中人脸识别率达到94.5%。
A method which can be used to identify the person of a video were proposed.First identify the faces of all the frames.Second calculate the proportion of the face in each frame,and the key frames can be found.Last use the SIFT algorithm to match the face storeroom with the key frames;According to the predetermined threshold,output the result.The algorithm shows： it don＇t need lots of videos to exercise and assure the accuracy of the human face identification system.It＇s easy to pick up the key frame and save time which support the real time of human face identification system.Among them,the face in video identification rate hits 94.5%.
出处
《黑龙江大学工程学报》
2012年第1期98-104,共7页
Journal of Engineering of Heilongjiang University
关键词
人脸识别系统
SIFT
关键帧
视频
human face identification system
SIFT
key frame
video
