| 黄莺.基于相关反馈的特征融合图像检索优化策略初探[J].数字图书馆论坛,2018,(2):45~51 |
| 基于相关反馈的特征融合图像检索优化策略初探 |
| Image Retrieval Based on Relevance Feedback |
| |
| DOI： |
| 中文关键词: 相关反馈;图像检索;特征融合;不同一性 |
| 英文关键词: Relevance Feedback;Image Retrieval;Multi-Feature Integration;Nonidentity |
| 基金项目:本研究得到西南民族大学2016年度中央高校基本科研业务费专项资金项目"智慧政府战略中政务数据开放机制与平台建设研究"(编号:2016NZYQN26)资助. |
|
| 摘要点击次数: 1924 |
| 全文下载次数: 1493 |
| 中文摘要: |
| 本文介绍基于相关反馈实现自动语义标注的多特征融合图像检索方法的原理和发展,深入分析该方法自身存在的局限,包括对用户查询关键词的重视不够、偏重旧图像、"过反馈"、训练样本非对称.基于以上不足,从用户体验出发,本文提出通过基于信息资源不同一性的特殊属性改进排序结果,并提出优化关键词权值的更新方法,并在相似度计算中考虑用户查询关键词权重.这些策略能让用户更快地获取满足需求的图像,并从而改善用户体验,缓解基于相关反馈的多特征融合图像检索方法的不足. |
| 英文摘要: |
| Introducing the principle of the multi-feature integration image retrieval based on the automatic semantic annotation by means of the relevance feedback, and summarizing the further development and the limitations, which are particular stress on the less emphasis on the query keywords the users input, the outdated images, over-feedback. Based on these limitations and aiming at improving user experience, the paper proposes a series of improvement methods, which respectively are a method to improve the result ranking is proposed based on the particular attribute-nonidentity of information recourse, the optimization of the updating the feature words' weights, weighting the keywords in the query. These strategies can improve the user experience, optimize the updating the feature words' weights and alleviate the problem of over-feedback and other preceding limitations. Especially, the result ranking based on the nonidentity can accelerate the access of the diverse desired images and improve the accuracy of the semantic web to describe the semantic feature of the images. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
