摘要：

针对目前视频人脸替换研究中对人脸位置的时间连续性和人脸检测的实时性要求较高的问题,提出一种基于帧间约束模型的视频人脸定量检测算法。利用前后帧的相关性建立帧间约束模型,定量描述视频人脸的具体位置,避免单帧检测的偏差,同时自适应地改变人脸搜索区 域以提高算法的运行速度。考虑到眉毛和眼睛的相似性,通过增加2种Haar_like EB特征,降低基于Haar_like特征人脸检测算法的误检数与漏检数。实验结果表明,该算法对视频中人脸位置的定量检测即时间连续性有所提升,且能够提高视频人脸检测的运行速率,降低误检率 。

关键词: 帧间约束模型, Haar_like EB特征, 定量检测, 人脸位置, 自适应搜索区域, 视频序列

Abstract:

Aiming at the problem that the temporal continuity of face position and the real-time requirement of face detection are high in the current video face replacement research,a video face quantitative detection algorithm based on the interframe constraint model is proposed.The interframe constraint model is established by using the correlations of the frames before and after,and the specific position of the video face is quantitatively described to avoid the deviation of single frame detection.At the same time,adaptively changing the face search area improves the running speed of the algorithm.Due to the similarity of eyebrows and eyes,the false detection and missed detection of face detection algorithms based on Haar_like features is reduced by adding two kinds of Haar_like EB features.Experimental results show that the algorithm can improve the temporal continuity of the quantitative detection of face position in video,and it can improve the running speed of video face detection and reduce the false detection rate.

Key words: interframe constraint model, Haar_like EB feature, quantitative detection, face location, adaptive search area, video sequences

中图分类号: 

• TP391