摘要
在时下的无监督深度哈希研究领域中,基于对比学习而提出的方法占主流地位.但是对比学习所采用的随机抽取负样本方式,会带来采样偏差问题,从而对图像检索精度造成负面影响.为解决以上问题,本文提出了一种基于偏差抑制对比学习的无监督深度哈希(unsupervised deep hashing based on bias suppressing contrastive learning,BSCDH).本文在对比学习框架的基础上提出了偏差抑制方法(bias suppression,BSS),将错误负样本近似为极困难负样本,并设计了偏差抑制系数来抑制极困难负样本从而缓解采样偏差的负面影响.本文根据当前负样本与查询样本的相似度来确定其对应的抑制系数取值,并引入当前负样本与邻近的聚类中心间的距离关系对抑制系数进行取值修正,降低正常负样本被过度抑制的可能性.最终BSCDH的64位哈希码mAP@5000指标在CIFAR-10、FLICKR25K、NUS-WIDE数据集上分别达到0.696、0.833、0.819,相较baseline具有显著的性能优势.本文开展的大量实验证明了BSCDH在无监督图像检索方法中拥有较高的检索精度,且能有效应对采样偏差问题.
In the contemporary field of unsupervised deep hashing research,methods predicated on contrastive learning are predominant.However,sampling bias brought about by the random extraction of negative samples in contrastive learning deteriorates image retrieval accuracy.To address the issue,this study proposes a novel unsupervised deep hashing based on bias suppressing contrastive learning(BSCDH).It proposes a bias suppression method(BSS)based on a contrastive learning framework.This method approximates incorrect negative samples as extremely hard negative samples and designs a bias suppression coefficient to suppress these extremely hard negative samples,thereby alleviating the negative impact of sampling bias.The corresponding suppression coefficient value is determined based on the similarity between the current negative sample and the query sample.Distance relationship between the current negative sample and adjacent hash centers is introduced to correct the suppression coefficient value,reducing the possibility of excessive suppression of normal negative samples.Ultimately,the mAP@5000 of the BSCDH method(64 bits)achieves 0.696,0.833,and 0.819 respectively on the CIFAR-10,FLICKR25K,and NUS-WIDE datasets,demonstrating a significant performance advantage over the baseline.Extensive experiments conducted in this paper verify that BSCDH exhibits high retrieval accuracy in unsupervised image retrieval methods and can effectively address sampling bias.
作者
苏海
钟雨辰
SU Hai;ZHONG Yu-Chen(School of Software,South China Normal University,Guangzhou 528225,China)
出处
《计算机系统应用》
2025年第2期165-173,共9页
Computer Systems & Applications
基金
广东省基础与应用基础研究基金(2022A1515140110,2021A1515110673,2020B1515120089)。
关键词
无监督哈希
对比学习
图像检索
采样偏差
unsupervised hashing
contrastive learning
image retrieval
sampling bias
