摘要
为了提高新闻视频单元分割的准确性和稳定性,提出了多特征融合的新闻视频单元高效分割技术。首先,通过基于重要性区域分析的局部直方图比较方法,实现新闻镜头的准确检测,并提取出每个镜头的关键帧。然后,检测出关键帧中的人脸信息,并针对局部直方图和人脸特征实现关键帧聚类,从而确定口播镜头的位置。最后,通过解析串联单进一步优化新闻单元分割的结果。在省、市、县播出的多个新闻视频上进行实验,结果表明该方法具有较高的可靠性和稳定性,能够准确高效地实现新闻单元的自动分割。
In order to improve the accuracy and stability of news unit segmentation, an efficient technique based on multiple feature fusion is proposed. Firstly, the shot boundary is correctly detected by the saliency-ware local histogram comparison, and keyframes are extracted to represent each shot. Then, human faces in each keyframe is detected, and keyframes based on local histogram and face similarities are clustered, accordingly the anchorperson shots can be located. Finally, news unit segmentation results are optimized by analyzing the news planned order. The proposed method is tested on many news videos produced by provincial, city and country level television station, and experimental results show that the method is reliable and stable, and can automatically segment news units correctly and efficiently.
出处
《电视技术》
北大核心
2016年第11期132-136,共5页
Video Engineering
基金
浙江省自然科学基金项目(LY14F020050
LQ14F020008)
浙江省科技计划项目(2015C33069)
浙江广播电视技术研究所2015年度科研项目(2015004)
关键词
多特征融合
新闻单元分割
聚类
人脸识别
串联单
multiple feature fusion
news unit segmentation
clustering
face recognition
news planned order