Abstract: A new method based on clustering algorithm is presented to detect anchorperson shot in news video. At first, representative frames of all shots are grouped by clustering based on face information. Then the color characteristics of persons and the border of persons are used to confirm anchorperson shot types. For consummating the anchorperson shot types, the method analyzes similarity of the background marginal of representative frames between anchorperson shot and others. Experimental results show that this method has high efficiency and adaptability.

Key words: news video, anchorperson shot, face character, clustering

摘要： 提出一种基于聚类算法的新闻视频播音员镜头探测方法。该方法利用人脸探测得到的人脸属性进行聚类，在此基础上利用人物及人物边缘背景区域的颜色特征在聚类的结果中进行聚类分析，确定播音员镜头候选类型。通过背景边缘的颜色相似度分析对播音员镜头候选类型进行完善得到最终的播音员镜头类型。实验证明该方法通用性和效率都比较好。

关键词: 新闻视频, 播音员镜头, 人脸属性, 聚类

CLC Number: 

• TP391.41