摘要
全媒体新闻中包含影像、字幕、语音、人物等多模态信息,已有的计算机编目系统常对单模态信息进行处理,生成的编目信息不够全面和准确。本文提出一种基于多模态融合的全媒体新闻智能编目系统,它通过多模态媒体内容统一表征与融合理解的核心算法来自动化生成拆条标记、影像描述、关键词、场景分类标签等关键信息,实现对新闻内容的全流程智能编目,提升了编目信息准确性并降低了人工成本。
Omnimedia news contains multimodal information ranging from video,subtitle,speech to character.Existing computer cataloging systems often process single-modal information,and generated cataloging information is not comprehensive and accurate.This paper proposed a multimodal fusion-based cataloging system for omnimedia news.The system uses core algorithm of unified representation and fusion understanding of multimodal media content to automatically realize shot segmentation,content description and comprehension,keywords generation,and scene classification.It realizes whole-process intelligent cataloging of news content,improves the accuracy of cataloging information and reduces labor costs.
作者
张随雨
俞定国
马小雨
钱永江
Zhang Suiyu;Yu Dingguo;Ma Xiaoyu;Qian Yongjiang(Communication University of Zhejiang,Zhejiang 310018,China;Zhejiang Radio and Television Group,Zhejiang 310012,China;Key Lab of Film and TV Media Technology of Zhejiang Province,Zhejiang 310018,China)
出处
《广播与电视技术》
2022年第2期62-65,共4页
Radio & TV Broadcast Engineering
基金
浙江省重点研发计划项目(2019C03138)资助。
关键词
多模态
视频编目
编目系统
深度学习
全媒体新闻
Multimodal
Video cataloging
Cataloging system
Deep learning
Omnimedia news
