短视频场景分类方法综述
A survey of micro-video scene classification
聂秀山 1巩蕊 1董飞 2郭杰 1马玉玲1
作者信息
- 1. 山东建筑大学计算机科学与技术学院,山东 济南 250101
- 2. 山东师范大学新闻与传播学院,山东 济南 250358
- 折叠
摘要
传统的视频场景分类方法习惯于从视觉模态中提取表现图像场景的特征,结合支持向量机等有监督学习方法,实现对某些类别的场景分类.随着各种短视频在各大平台迅速涌现,基于短视频特性的场景特征表示越来越受到研究者们的关注.由于短视频数据具有噪声、数据缺失、各模态语义强度不一致等问题,导致传统的视频场景表征方法无法学习具有丰富语义的短视频场景表征.近年来,部分短视频场景分类的研究考虑上述挑战,并提出相应的方法.本研究综述短视频场景分类的研究现状,介绍短视频场景特征表示和分类方法,对不同数据集上的场景分类方法进行分析.针对现有方法存在的问题,分析未来短视频场景分类中需要解决的挑战性问题.
Abstract
Traditional video scene classification methods were used to extract the features of image scenes from the visual modality,and combined with supervised learning methods such as support vector machine to achieve scene classification of certain categories.With the rapid emergence of various micro-videos on major platforms,the scene feature representation based on the characteristics of micro-videos had attracted more and more attention of researchers.Due to the problems of micro-video data such as noise,data loss,and inconsistent semantic intensity of each modality,these issues resulted in traditional methods for representing video scenes being unable to learn micro-video scene representations with rich semantics.In recent years,the research of some micro-video scene classi-fication had considered the above challenges and proposed corresponding methods based on micro-video scene classification.This study reviewed the research status of micro-video scene classification,introduced the feature representation and classification methods of micro-video scene,and analyzed the scene classification methods on different datasets.Aiming at the problems existing in the existing methods,the challenging problems to be solved in the future micro-video scene classification were analyzed.
关键词
视频场景/特征表示/短视频场景分类/多模态融合/深度学习Key words
video scene/feature representation/micro-video scene classification/multi-modality fusion/deep learning引用本文复制引用
基金项目
国家自然科学基金(62176141)
国家自然科学基金(62176139)
国家自然科学基金(61876098)
山东省杰出青年自然科学基金(ZR2021JQ26)
山东省自然科学基金(ZR2021QF119)
出版年
2024
NETL
NSTL
万方数据