Multi-Modal Multiple-Instance Learning and Attribute Discovery with the Application to the Web Violent Video Detection
Multi-Modal Multiple-Instance Learning and Attribute Discovery with the Application to the Web Violent Video Detection
复制标题
DOI：
10.1007/978-3-642-42057-3_57
复制
发表时间：
2013-07
影响因子：
4.7
通讯作者：
Shuai Hao;Ou Wu;Weiming Hu;Jinfeng Yang
中科院分区：
文献类型：
--
作者：
Shuai Hao;Ou Wu;Weiming Hu;Jinfeng Yang
沿着互联网的不断发展，网络暴力视频的传播干扰了我们的日常生活，影响了我们，特别是儿童的健康。因此，暴力视频识别对于网络内容过滤变得越来越重要。本文将音视频与文本信息相结合，利用多模态多实例学习和属性发现方法将视频分为暴力和非暴力两类进行Web视频检测。主要工作有两方面。首先，我们从视频数据中构建训练包，并使用属性学习来解释属性关系。其次，我们设计了一个有效的实例选择技术，利用音频，视频和文本信息，以加快训练过程中，不影响性能。对200个视频共享网站的实验结果表明，该方法对暴力视频的检测是有效的。
Along with the ever-growing web, violent video sharing in the Internet has interfered with our daily life and affected our, especially children’s health. Therefore violent video recognition is becoming important for web content filtering. In this paper, we classified the video into violent and nonviolent using Multi-Modal Multiple-Instance Learning and Attribute Discovery approach by combining audio-video with text information for web video detection. The main work is two-fold. First, we build the training bags from our video data and use attribute learning to explain attributes relation. Second, we design an efficient instance selection technique by utilizing audio-video and text information to speed up the training process without compromising the performance. The experimental results on 200 videos collected from the sharing video sites show that the proposed method is effective on violent video detection.