摘要
网络的快速发展在给人们生活带来很多便利的同时,也存在着令人反感的敏感内容,如恐怖暴力视频等,这些内容严重影响着青少年的身心健康.因此,一种有效的敏感视频识别算法成为网络过滤技术中不可或缺的组成部分.近年来,多示例学习被引入到恐怖暴力类敏感视频识别中,并取得了令人瞩目的效果.由于该类视频中存在着很多冗余信息及部分非恐怖暴力帧的干扰,不可避免地影响了敏感视频的识别效果.因此提出了一种基于判别性特征投影的多示例学习算法,提出了一种基于自表示字典学习的示例选择框架,将示例作为字典,学习示例之间的最优表达关系,找到具有代表性的示例,并向代表示例进行投影构造了更具判别性的示例包特征.通过在恐怖暴力视频库以及VSD2014数据集上与现有多示例检测算法在准确率、召回率以及F1指标的对比,验证了该算法在恐怖暴力视频识别中的有效性.
Along with the convenience of Internet,there are also objectionable contents,e.g.horror and violent video which would seriously affects children’s physical and psychological health.So,an effective objectionable video recognition algorithm is necessary for web filtering.Recently,the multi-instance learning(MIL)has been introduced for horror and violent video recognition and achieved impressed results.However,many non-objectionable frames and noises in objectionable video scene is an intractable problem that inevitably degrades the recognition performance.In this paper,we propose a new method based on MIL via discriminative bag feature mapping.In the proposed method,an instance selection framework based on self-expressive dictionary learning is designed in which all instances are chosen as dictionary atoms and the representative instances can be selected by dictionary learning.Then bag level feature is constructed via the selected representative instances.Experiments on horror and violent video data sets show that our method is effective and efficient for objectionable scene recognition.
作者
丁昕苗
郭文
吕文恬
Ding Xinmiao;Guo Wen;Lyu Wentian(School of Information and Electronic Engineering,Shandong Technology and Business University,Yantai 264005)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2020年第5期804-810,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61876100,61572296,61801269)
山东省高等学校青创科技支持计划(2019KJN041).
关键词
敏感视频
自表示字典学习
视频情感
网络过滤
objectionable video
self-express dictionary learning
video affection
web filter
