...
  • 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>電子情報通信学会技術研究報告. パターン認識·メディア理解. Pattern Recognition and Media Understanding >Semantic Video Concept Detection using Subspace-partition based Scheme - TRECVid 2012 Semantic Video Concept Detection by NTT
【24h】

Semantic Video Concept Detection using Subspace-partition based Scheme - TRECVid 2012 Semantic Video Concept Detection by NTT

机译：使用基于子空间分区的方案进行语义视频概念检测-TRECVid 2012 NTT进行语义视频概念检测

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In this paper, we describe the TRECVid 2012 video concept detection system. For this year's task, we adopted a subspace -partition based scheme for classifier learning, which puts emphasis on the reduction of classifier complexity and aims at improving the training efficiency and boosting the classifier performance. As the video corpus used for TRECVid evaluation is ever increasing, two practical issues are becoming more and more challenging for building concept detection systems. The first one is the time-consuming training and testing procedures, which have taken up most of the evaluation activities, preventing the design and testing of novel algorithms. The second and the more important issue is that when using whole data for classifier training, the derived separating hyperplanes would be rather complex and thus degrade the classification performance. To address these issues, we propose to adopt the "divide-and-conquer" strategy for concept detector construction as follows. We first partition the whole training feature space into multiple sub-space with a scalable clustering method, and then build sub-classifiers on these sub-spaces separately for each concept. The decision of a testing sample is the fusion of the results a few fired sub-classifiers. Experimental results demonstrate the efficiency and effectiveness of our proposed approach.
机译：在本文中，我们描述了TRECVid 2012视频概念检测系统。对于今年的任务，我们采用了基于子空间分区的分类器学习方案，该方案着重于降低分类器复杂度，并旨在提高训练效率和提高分类器性能。随着用于TRECVid评估的视频语料库不断增加，对于构建概念检测系统，两个实际问题变得越来越具有挑战性。第一个是耗时的培训和测试过程，该过程占用了大多数评估活动，从而阻碍了新算法的设计和测试。第二个也是更重要的问题是，当使用整体数据进行分类器训练时，派生的分离超平面会相当复杂，从而降低分类性能。为了解决这些问题，我们建议采用“分而治之”的策略来构造概念检测器，如下所示。我们首先使用可扩展的聚类方法将整个训练特征空间划分为多个子空间，然后针对每个概念在这些子空间上分别构建子分类器。测试样本的决定是将结果合并到几个发射的子分类器中。实验结果证明了我们提出的方法的效率和有效性。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号