基于人脸检测与SIFT的播音员镜头检测

播音员镜头的检测是新闻视频结构化的关键步骤之一.提出了一种基于人脸检测与SIFT特征点匹配的播音员镜头自动检测算法.该方法首先利用人脸检测器过滤出具有人脸的候选镜头,然后利用颜色直方图判断镜头是否可能相似,再利用SIFT特征点匹配从候选镜头关键帧中找出相关的镜头组,最后利用各镜头组的信息判断出哪些是播音员镜头.对比传统的方法,该方法除了训练一个通用的人脸检测器外,不需要模板,也不需要针对某类新闻节目训练特别的分类器,可以直接利用算法对新类型的新闻节目提取播音员镜头.实验结果表明,该算法能够广泛地适应于各种不同种类的新闻节目、不同视觉质量的视频,可以有效地应用于新闻视频分析.

Anchor Shot Detection Based on Face Detection and SIFT

Anchor shot detection is a fundamental step for segmenting news video into stories. In this paper, a algorithm is developed for anchor shot detection based on face detection and SIFT. Face detection is first used to filter out the shots which do not have any face in the special area. Then, color histogram is used to judge whether two shots are similar, and the distinctive image features from scale-invariant key points are detected and matched to find groups of shots which may be anchor shots. Last, anchor shots are identified based on the same prior information of anchor shots. Compared with other algorithms, the significant advantage of the proposed algorithm is that the algorithm is based on neither the templates nor the learned classifiers. The method has been tested on many kinds of news video, which demonstrates its effectiveness.