摘要
针对新闻图像检索的应用特点,提出了一种多反馈、合作型的图像检索方法。通过构造动态的Q表,保存图像的折算累计反馈;设计从探索型逐渐过渡到利用型的图像选择策略;在方差分析的基础上,设计了多反馈综合方法,全面地获取用户检索需求,从而构造了基于Q学习的相关反馈检索算法。实验结果表明了该算法是有效的,并具有更高的性能。
According to application features of news image retrieval, a multi-feedback and cooperative retrieval method is pro posed. A dynamic Q-chart is constructed to store discounted cumulative rewards of images. Image selecting strategy is designed transited from exploring to exploiting. And, a multi-feedbacks synthesizing method is constructed at the basis of variance analysis to gain total requirement of user. So, a relevance feedback algorithm based on Q-learning is proposed. Experimental results show the proposed algorithm is effective and has higher performance.
出处
《计算机工程与设计》
CSCD
北大核心
2012年第8期3210-3213,共4页
Computer Engineering and Design
基金
天津师范大学教育科学研究基金项目(52WT1114)
天津市应用基础及前沿技术研究计划基金项目(10JCYBJC26600)
关键词
图像检索
相关反馈
Q学习
选择策略
折算累计反馈
image retrieval
relevance feedback
Q-learning
selecting strategy
discounted cumulative reward
