从视频中检测人脸

视频中人脸检测的应用领域广泛，近来受到了极大关注。文中提出一种在MPEG流中检测人脸的新方法，它可以从复杂背景中有效地检测方向、大小不同的人脸，还可以处理多个人脸交叠的情况，为适应视频检索的需要，该算法依据帧间冗余性，自适应地调整肤色检测器，利用MPEG流中的运动矢量在一个GOP内跟踪人脸，依据场景的变化更新分割码本等措施，有效地提高了计算速度。用算法测试多个视频序列，实验结果令人满意。

Detect Human Faces in Video

Face detection in video has received considerable attention in recent years due to numerous possible applications. A new algorithm that can detect multi faces with multiple orientation and scales against a complex background in MPEG video sequences is presented, which can also detect faces from where multi faces overlap with each other. To satisfy the need of video retrieval, several means are applied to save the computation cost. These means include tuning the skin filter according to the redundancy among frames, tracking faces in a GOP using motion vectors directly from the MPEG format, updating the code book for segmentation according to the change of scene. New algorithm is applied to multiple test video sequence, and results are promising.