基于监督学习的跨模态图文检索方法  

Cross-Modal Image and Text Retrieval Based on Supervised Learning

作　　者：赵晓峰[1,2] 张东霖 吴小俊 ZHAO Xiao-feng;ZHANG Dong-lin;WU Xiao-jun(School of I0T&AI,Wuxi Vocational Institute of Commerce,Wuxi Jiangsu 214064,China;School of AI&CS,Jiangnan University,Wuxi Jiangsu 214122,China)

机构地区：[1]无锡商业职业技术学院物联网与人工智能学院,江苏无锡214064 [2]江南大学人工智能与计算机学院,江苏无锡214122

出　　处：《计算机仿真》2024年第10期517-523,共7页Computer Simulation

摘　　要：近年来,由于哈希技术在计算速度和存储成本方面的优势,使得跨模态哈希在信息检索领域获得了广泛的关注。然而现有大多数方法只是简单地探索全局相似性,而多模态数据中的局部信息常被忽略。另外现有的大多数方法常采用一步策略同时哈希码和哈希函数,常会导致优化难解决并增加计算复杂度。为此提出了一种新的算法,即鉴别性监督跨模态哈希方法。方法联合考虑全局和局部信息并将其嵌入到哈希学习中并设计了一个跨模态哈希函数学习方案以学习强大的哈希函数。在两个大规模数据集上进行跨模态检索实验,实验结果表明,所提方法可以获得良好的检索性能,验证了该算法的有效性。In recent years,recent years,due to the advantages of hash technology in computing speed and storage cost,cross modal hashing has received widespread attention in the field of information retrieval.However,most existing methods simply explore global similarity,while local information in multimodal data is often overlooked.In addition,most existing methods often adopt a one-step strategy to simultaneously hash codes and hash functions,which often leads to difficult optimization and increases computational complexity.To this end,we develop a novel CMH algorithm,namely,Discriminative Supervised Cross-modal Hashing.The global and local information is jointly considered and embedded into hash code learning.Moreover,to learn powerful hash function,we deliberately develop a cross-modal hash function learning scheme.To verify the efficacy of the proposed model,we perform some crossmodal search experiments on two large-scale datasets.The results demonstrate that the developed method can achieve good retrieval performance,verifying the efficacy of the proposed method.

关 键 词：跨模态 哈希 相似性检索 监督 离散约束 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]