摘要
本文提出了一种基于贝叶斯学习的视频检索相关反馈算法,该算法使用慨率框架描述检索问题,根据贝叶斯公式按照用户标记相关和不相关样本集来更新视频库中镜头的目标概率,实现自动相关反馈。通过增量学习,使系统的检索能力不断得到提高。对包含几千个镜头的视频库的实验表明,新算法能够显著提高检索的性能,并在有限的几次反馈后就能快速收敛于用户的查询概念。
An algorithm of video retrieval based on Bayesian learning relevance feedback was presented. It uses a probabilistic framework to describe retrieval problem. The system updates the probability distribution automatically via Bayesian formulation according to user action. The results of experiment on thousands of shots show that the proposed approach can improve the efficiency and accuracy of the search performance and grasp user＇ s query concept only several iterations.
出处
《信号处理》
CSCD
北大核心
2009年第3期408-411,共4页
Journal of Signal Processing
关键词
视频检索
相关反馈
贝叶斯学习
video retrieval
relevance feedback
Bayesian learning
