计算机应用研究2026，Vol.43Issue(3)：924-930,7.DOI:10.19734/j.issn.1001-3695.2025.05.0226
基于非对称框架的高效图像语义检索范式
Efficient semantic image retrieval paradigm via asymmetric framework
摘要
Abstract
This study proposed OnlyGlobal,an asymmetric global feature retrieval paradigm,to address the computational re-dundancy and poor scalability inherent in the two-stage global feature matching and re-ranking framework for image retrieval.It designed a collaborative optimization framework that integrated multi-modal feature fusion on the offline gallery side with light-weight processing on the online query side,transforming the advantages of traditional re-ranking into offline feature enhance-ment.This method firstly constructed a multi-granularity feature space by applying multi-scale transformation and semantic seg-mentation to gallery images offline.It then introduced a keypoint heatmap-guided feature enhancement mechanism to strengthen discriminative region representation by suppressing background noise.Finally,it generated enhanced gallery features using adaptive feature aggregation,establishing a dimensionally consistent asymmetric matching model.Experiments demonstrate that this approach achieves significant performance gains over state-of-the-art methods while maintaining the simplicity of a single-stage architecture.By eliminating online multi-scale processing and re-ranking computation,this method improves query feature extraction efficiency by over 300％and achieves order-of-magnitude speedup in matching.The work provides a viable technical pathway for deploying billion-scale image retrieval systems.关键词
图像检索/非对称检索/语义分割/关键点检测Key words
image retrieval/asymmetric retrieval/semantic segmentation/keypoint detection分类
信息技术与安全科学引用本文复制引用
王越芸,黄华..基于非对称框架的高效图像语义检索范式[J].计算机应用研究,2026,43(3):924-930,7.基金项目
2024年上海开放大学新进人员科研扶持项目(XJ2404) （XJ2404）
