基于边界归类的新闻视频故事分割算法  

News Story Segmentation Based on Boundary Classification

作　　者：王娟[1] 蒋兴浩[1,2] 孙锬锋[1,2] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240 [2]上海交通大学信息内容分析技术国家工程实验室,上海200240

出　　处：《上海交通大学学报》2016年第9期1384-1389,1398,共7页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金项目(61272439;61272239;61572320;61572321);教育部博士点专项基金项目(20120073110053)

摘　　要：为有效组织和浏览新闻视频,提出了一种基于边界归类的新闻视频故事分割算法.该算法视镜头边界为候选故事边界,并定义新闻基本处理单元划分新闻视频.算法分为新闻基本处理单元的获取和新闻基本处理单元分析两部分.前者采用镜头分割和镜头标定对原始视频进行合理划分,获取基本处理单元边界,有效缩小故事边界判定范围;后者对基本处理单元内的字幕文本进行分析,实现了字幕文本分类和主题字幕相似性比较,并结合静音特征,从音视频两方面判定故事边界,得到最终的分割结果.实验结果表明,该算法能有效描述新闻故事边界,准确分割新闻故事单元,实现对新闻视频的语义划分,为新闻视频检索、导航等应用提供前期辅助.In order to organize and browse news video, a story segmentation scheme based on boundary classification was proposed. The proposed scheme regarded shot boundaries as candidate story boundaries, and defined news basic unit to divide the news video. The scheme can be divided into news basic unit gain and news basic unit analysis. The former narrows the story boundaries by using shot segmentation and shot demarcation while the latter gains the final segmentation result by analyzing the caption in news, achieving the classification of caption and comparison of topic caption, by combining with silence feature. The experimental results show that the proposed scheme can express story boundaries effectively, divide the news into story unit precisely and realize news video semantic partition, which can provide convenience to news retrieval and news indexing.

关 键 词：新闻故事分割 新闻基本处理单元 镜头边界 主题字幕 静音检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]