Article(id=1322961012428775513, tenantId=1146029695717560320, journalId=1249024232475115590, issueId=1322960856543273127, articleNumber=null, orderNo=null, doi=10.11834/jig.250393, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1761580800000, receivedDateStr=2025-10-28, revisedDate=1768406400000, revisedDateStr=2026-01-15, acceptedDate=null, acceptedDateStr=null, onlineDate=1793348084789, onlineDateStr=2026-10-30, pubDate=1786809600000, pubDateStr=2026-08-16, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1793348084789, onlineIssueDateStr=2026-10-30, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1793348084789, creator=13701087609, updateTime=1793348084789, updator=13701087609, issue=Issue{id=1322960856543273127, tenantId=1146029695717560320, journalId=1249024232475115590, year='2026', volume='31', issue='8', pageStart='2687', pageEnd='3074', issueExtLink='null', onlineDate='null', pubDate='1786809600000', pubDateStr='2026-08-16', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1793348047624, creator='13701087609', updateTime=1793348605344, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext={EN=IssueExt(id=1322963195874378645, tenantId=1146029695717560320, journalId=1249024232475115590, issueId=1322960856543273127, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1322963195874378646, tenantId=1146029695717560320, journalId=1249024232475115590, issueId=1322960856543273127, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null, downloadFileDto=null}, startPage=3030, endPage=3044, ext={EN=ArticleExt(id=1322961012625907802, articleId=1322961012428775513, tenantId=1146029695717560320, journalId=1249024232475115590, language=EN, title=Image-text retrieval for Han portrait stones through the synergy of semantic enhancement and text reorganization, columnId=1249044008786400014, journalTitle=Journal of Image and Graphics, columnName=Image Understanding and Computer Vision, runingTitle=null, highlight=null, articleAbstract=
Objective

Han portrait stones are decorative stone carvings on funerary architectural structures,such as tombs and ancestral halls,dating back to the Han Dynasty. The contents of these stones depict mythological legends,institutional systems,and everyday life scenes,including chariot processions,kitchen banquets,textile farming,and other scenes that directly reflect the social practices of the Han period. Iconic figures,such as the Queen Mother of the West (Xiwangmu),Fuxi and Nüwa,and feathered immortals (Yuren),represent the Han people’s cosmology of “immortal souls”. In particular,Xiwangmu images are prevalent in the tombs of the middle and lower classes but seldom appear in noble tombs,reflecting hierarchical differences in beliefs. Themes,such as ancient sages and war scenes,which combine architectural practicality and artistic decoration,serve educational and commemorative functions,embodying realism and romanticism. This study proposes an image-text retrieval method that leverages semantic information from mask images of Han portrait stones to enhance images and texts,addressing the domain shift problem present in existing image-text retrieval methods. The method introduces a semantic enhancement module and integrates semantic information with randomly generated prompt templates for text reorganization. This design enables the model to learn additional discriminative features,understand different objects in images,align text with images,and improve the accuracy and comprehensiveness of image-text retrieval.

Method

This study utilizes the architecture of Chinese contrastive language-image pre-training (Chinese CLIP) using a vision transformer (ViT) as the image encoder and robustly optimized bidirectional encoder representations from Transformers (RoBERTa) model as the text encoder to extract image and text features. Semantic information from mask images and original text is used to calculate object-specific weights. Attention weights are derived by combining these weights with features from mask and original images. After a residual connection,these attention weights are used to generate attention features,while the weights themselves serve as semantic enhancement features. The original image features,attention features,and enhancement features are combined through a multi-stage feature fusion strategy at a ratio of 0.4:0.5:0.1 to form composite features. This design enables the model to focus on the primary subjects and important objects in images while enriching feature representation without information loss. Effective semantic objects are extracted from the mask images,and the original texts are embedded into randomly generated prompt templates incorporating these objects to produce enhanced texts. This process structures text descriptions,strengthens semantic expression,and improves matching with image features. Furthermore,dynamic temperature adjustment and asymmetric similarity calculation are introduced: the contrastive learning temperature is adaptively adjusted based on feature similarity (low temperatures applied to high-similarity samples to enhance discriminative learning and high temperatures applied to low-similarity samples to increase tolerance). The different temperature coefficients (0.9 for image-to-text and 1.1 for text-to-image) balance the matching difficulty in both directions,accounting for the distribution differences between image and text features to improve the accuracy and stability of cross-modal matching.

Result

In this study,a series of comparative experiments,ablation experiments,and visual comparison experiments is conducted on the proposed model to evaluate the model’s performance and the contributions of each module. Results show that,on the Han portrait stone test set,the proposed method significantly improves the text-to-image retrieval metrics Mean_Recall (%),Recall@1 (%),Recall@5 (%),and Recall@10 (%) by 15.20%,21.84%,14.29%,and 9.48%,respectively,compared with the second-ranked bootstrapping language-image pre-training model in terms of precision and comprehensiveness. During the image-to-text retrieval,the metrics Mean_Recall (%),Recall@1 (%),Recall@5 (%),and Recall@10 (%) are improved by 17.47%,23.22%,18.45%,and 10.72%,respectively,compared with the second-ranked cross-modal vision-language model (X2-VLM),demonstrating significant improvements in image-text retrieval performance. Visualization experiments can effectively assist us in observing the improvement of the model’s performance and the enhanced effects brought about by each added module,making the experimental results highly intuitive. Additionally,the proposed method maintains good retrieval performance in data robustness experiments. The average recall rate of the model consistently exceeds 70%,and it maintains excellent retrieval performance. This outcome indicates that the proposed model exhibits remarkable data robustness and is applicable to various scenarios,such as fuzzy retrieval.

Conclusion

The proposed method enhances the model’s ability to align Han portrait stone images with their descriptive texts. The model can learn effective and crucial features by simultaneously enhancing text and images. The proposed dynamic temperature mechanism adapts to the structural characteristics of the Han portrait stone dataset and enhances the model’s ability to learn from difficult samples. effectively improving the performance of image-text retrieval in terms of accuracy and comprehensiveness. However,when objects in images are similar but differ in positional information,the image-text retrieval results may suffer from confusion and false positives. Given the limited scale and diversity of the dataset for Han portrait stone image-text retrieval,the model’s cross-modal retrieval performance may be constrained. In the future,we will optimize the model’s capability to semantically analyze the different positions and actions of objects in images and expand the Han portrait stone dataset.

, authors=Kun Jiang, Wenhua Qian*, Peng Liu, authorsList=Kun Jiang, Wenhua Qian, Peng Liu, authorCompany=null, correspAuthors=Wenhua Qian, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, fund=null), CN=ArticleExt(id=1322961017071870069, articleId=1322961012428775513, tenantId=1146029695717560320, journalId=1249024232475115590, language=CN, title=语义增强与文本重组协同的汉画像石图文检索, columnId=1249044009054835474, journalTitle=中国图象图形学报, columnName=图像理解和计算机视觉, runingTitle=null, highlight=null, articleAbstract=
目的

汉画像石是解读汉代历史文化的珍贵载体。图文检索技术能推动汉代文化研究向数字化转型,为文化保护提供智能化解决方案。然而,现有图文检索方法应用于汉画像石时,存在领域偏移、图像与文本难以对齐等问题。本文提出一种语义增强与文本重组协同的图文检索方法以提高检索的准确性和全面性。

方法

本文通过编码器分别完成汉画像石图像与文本特征的提取。在图像处理阶段,基于汉画像石图像对应掩膜图像,计算各语义对象权重,进而推导得出注意力权重。通过残差连接获取注意力特征,并将所得注意力权重直接作为语义增强特征。通过多阶段特征融合将原始特征、注意力特征和增强特征融合得到组合特征以丰富特征表示。在文本处理阶段,将文本与提取的图像语义对象嵌入随机提示模板,生成结构更清晰的增强文本,增强文本的语义表达能力。同时,为适配本文数据集结构,提出动态温度调整以及非对称相似度计算的方法,基于特征相似度自适应调整温度。对图像到文本和文本到图像双向的相似度矩阵计算采用不同温度系数以提升跨模态匹配的准确性与稳定性。

结果

通过本文所提方法验证,文本到图像的检索指标Mean_Recall、Recall@1、Recall@5和Recall@10相较于排名第2的BLIP(bootstrapping language-image pre-training)模型分别提高了15.20%、21.84%、14.29%和9.48%。图像到文本的检索指标Mean_Recall、Recall@1、Recall@5和Recall@10相较于排名第2的X2-VLM(all-in-one pre-trained model for vision-language tasks)模型分别提高了17.47%、23.22%、18.45%和10.72%,汉画像石图文可视化检索也得到了更好的效果。

结论

本文方法显著增强了模型对汉画像石图像与其描述文本的对齐能力,有效提升了图文检索在准确性、全面性等方面的表现。

, authors=姜坤, 钱文华*, 刘朋, authorsList=姜坤, 钱文华, 刘朋, authorCompany=null, correspAuthors=钱文华, authorNote=

姜坤,男,硕士研究生,主要研究方向为深度学习和计算机视觉。E-mail:

钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:

刘朋,男,博士研究生,主要研究方向为计算机视觉和图像处理。E-mail:

, correspAuthorsNote=
钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=OkWF19JmSjOPh/esjasB0A==, magXml=ztubDT9MBLaz2zelvGA9Kg==, pdfUrl=null, pdf=fhmCK3oxCAjwjAuIHpNPgw==, pdfFileSize=8318651, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=lYnp/pVqgXO1dC7xP8MsTg==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=hK86WDzRphA163nfsdG7nw==, mapNumber=null, fund=null)}, authors=[Author(id=1322961017378054266, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=2476270892@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322961017453551740, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961017378054266, language=EN, stringName=Kun Jiang, firstName=Kun, middleName=null, lastName=Jiang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Information Science and Engineering,Yunnan University,Kunming 650504,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322961017533243517, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961017378054266, language=CN, stringName=姜坤, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=云南大学信息学院,昆明 650504, bio={"content":"

姜坤,男,硕士研究生,主要研究方向为深度学习和计算机视觉。E-mail:

"}, bioImg=null, bioContent=

姜坤,男,硕士研究生,主要研究方向为深度学习和计算机视觉。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322961017294168182, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, xref=null, ext=[AuthorCompanyExt(id=1322961017302556791, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Information Science and Engineering,Yunnan University,Kunming 650504,China), AuthorCompanyExt(id=1322961017310945400, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=云南大学信息学院,昆明 650504)])]), Author(id=1322961017612935295, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=whqian@ynu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1322961017730375809, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961017612935295, language=EN, stringName=Wenhua Qian, firstName=Wenhua, middleName=null, lastName=Qian, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=School of Information Science and Engineering,Yunnan University,Kunming 650504,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322961017784901762, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961017612935295, language=CN, stringName=钱文华, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=云南大学信息学院,昆明 650504, bio={"content":"

钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:

"}, bioImg=null, bioContent=

钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322961017294168182, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, xref=null, ext=[AuthorCompanyExt(id=1322961017302556791, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Information Science and Engineering,Yunnan University,Kunming 650504,China), AuthorCompanyExt(id=1322961017310945400, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=云南大学信息学院,昆明 650504)])]), Author(id=1322961019466817668, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=pengliu0606@gmail.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322961019563286662, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961019466817668, language=EN, stringName=Peng Liu, firstName=Peng, middleName=null, lastName=Liu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Information Science and Engineering,Yunnan University,Kunming 650504,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322961019638784135, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961019466817668, language=CN, stringName=刘朋, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=云南大学信息学院,昆明 650504, bio={"content":"

刘朋,男,博士研究生,主要研究方向为计算机视觉和图像处理。E-mail:

"}, bioImg=null, bioContent=

刘朋,男,博士研究生,主要研究方向为计算机视觉和图像处理。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322961017294168182, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, xref=null, ext=[AuthorCompanyExt(id=1322961017302556791, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Information Science and Engineering,Yunnan University,Kunming 650504,China), AuthorCompanyExt(id=1322961017310945400, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=云南大学信息学院,昆明 650504)])])], keywords=[Keyword(id=1322961019722670216, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, orderNo=1, keyword=Han portrait stones), Keyword(id=1322961019785584777, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, orderNo=2, keyword=image-text retrieval), Keyword(id=1322961019852693642, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, orderNo=3, keyword=semantic enhancement), Keyword(id=1322961019932385419, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, orderNo=4, keyword=text reorganization), Keyword(id=1322961019999494284, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, orderNo=5, keyword=dynamic temperature), Keyword(id=1322961020074991757, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, orderNo=1, keyword=汉画像石), Keyword(id=1322961020133712014, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, orderNo=2, keyword=图文检索), Keyword(id=1322961020192432271, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, orderNo=3, keyword=语义增强), Keyword(id=1322961020356010128, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, orderNo=4, keyword=文本重组), Keyword(id=1322961020511199377, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, orderNo=5, keyword=动态温度)], refs=[Reference(id=1322961025686970550, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2023, volume=45, issue=6, pageStart=7270, pageEnd=7292, url=null, language=null, rfNumber=null, rfOrder=0, authorNames=Chen W, Liu Y, Wang W P, Bakker E M, Georgiou T, Fieguth P, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Chen W,Liu Y,Wang W P,Bakker E M,Georgiou T,Fieguth P,et al. 2023. Deep learning for instance retrieval: a survey. IEEE Transactions on Pattern Analysis and Machine Intelligence,45(6): 7270-7292 [DOI: 10.1109/TPAMI.2022.3218591], articleTitle=Deep learning for instance retrieval: a survey, refAbstract=null), Reference(id=1322961025770856631, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=16, pageEnd=16, url=null, language=null, rfNumber=null, rfOrder=1, authorNames=Dosovitskiy A, Beyer L, Kolesnikov A, Weissenborn D, Zhai X H, Unterthiner T, journalName=An image is worth, refType=null, unstructuredReference=Dosovitskiy A,Beyer L,Kolesnikov A,Weissenborn D,Zhai X H,Unterthiner T,et al. 2021. An image is worth 16 × 16 words: transformers for image recognition at scale//Proceedings of the 9th International Conference on Learning Representations. [s.l.]: #ICLR, articleTitle=words: transformers for image recognition at scale, refAbstract=null), Reference(id=1322961025842159800, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=#12, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=2, authorNames=Faghri F, Fleet D J, Kiros J R, Fidler S, journalName=null, refType=null, unstructuredReference=Faghri F,Fleet D J,Kiros J R and Fidler S. 2018. VSE++: improving visual-semantic embeddings with hard negatives//Proceedings of 2018 British Machine Vision Conference. Newcastle,UK: BMVA Press: #12, articleTitle=VSE++: improving visual-semantic embeddings with hard negatives, refAbstract=null), Reference(id=1322961025921851577, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=3, authorNames=Gan S N, journalName=Study on the Dance Art in the Stone Relief of the Han Dynasty in Nanyang from the Perspective of Imagolog, refType=null, unstructuredReference=Gan S N. 2019. Study on the Dance Art in the Stone Relief of the Han Dynasty in Nanyang from the Perspective of Imagolog. Zhengzhou: Zhengzhou University, articleTitle=null, refAbstract=null), Reference(id=1322961026001543354, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=4, authorNames=甘胜楠, journalName=图像学视域下南阳汉画像石中的舞蹈艺术研究, refType=null, unstructuredReference=甘胜楠. 2019. 图像学视域下南阳汉画像石中的舞蹈艺术研究. 郑州: 郑州大学, articleTitle=null, refAbstract=null), Reference(id=1322961026085429435, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2024, volume=61, issue=4, pageStart=null, pageEnd=103716, url=null, language=null, rfNumber=null, rfOrder=5, authorNames=Ge X R, Xu S P, Chen F H, Wang J, Wang G X, An S, journalName=Information Processing and Management, refType=null, unstructuredReference=Ge X R,Xu S P,Chen F H,Wang J,Wang G X,An S,et al. 2024. 3SHNet: boosting image-sentence retrieval via visual semantic-spatial self-highlighting. Information Processing and Management,61(4): #103716 [DOI: 10.1016/j.ipm.2024.103716], articleTitle=3SHNet: boosting image-sentence retrieval via visual semantic-spatial self-highlighting, refAbstract=null), Reference(id=1322961026181898428, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=1999, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=6, authorNames=Hawking D, Voorhees E M, Craswell N, Bailey P, journalName=null, refType=null, unstructuredReference=Hawking D,Voorhees E M,Craswell N and Bailey P. 1999. Overview of the TREC-8 web track//Proceedings of 1999 Eighth Text Retrieval Conference. Gaithersburg,USA: NIST, articleTitle=Overview of the TREC-8 web track, refAbstract=null), Reference(id=1322961026253201597, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=770, pageEnd=778, url=null, language=null, rfNumber=null, rfOrder=7, authorNames=He K M, Zhang X Y, Ren S Q, Sun J, journalName=null, refType=null, unstructuredReference=He K M,Zhang X Y,Ren S Q and Sun J. 2016. Deep residual learning for image recognition//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas,USA: IEEE: 770-778 [DOI: 10.1109/CVPR.2016.90], articleTitle=Deep residual learning for image recognition, refAbstract=null), Reference(id=1322961026324504766, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=8, authorNames=Jang E, Gu S X, Poole B, journalName=null, refType=null, unstructuredReference=Jang E,Gu S X and Poole B. 2017. Categorical reparameterization with gumbel-softmax//Proceedings of the 5th International Conference on Learning Representations. Toulon,France: ICLR, articleTitle=Categorical reparameterization with gumbel-softmax, refAbstract=null), Reference(id=1322961026429362367, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=4904, pageEnd=4916, url=null, language=null, rfNumber=null, rfOrder=9, authorNames=Jia C, Yang Y F, Xia Y, Chen Y T, Parekh Z, Pham H, journalName=JMLR, refType=null, unstructuredReference=Jia C,Yang Y F,Xia Y,Chen Y T,Parekh Z,Pham H,et al. 2021. Scaling up visual and vision-language representation learning with noisy text supervision//Proceedings of the 38th International Conference on Machine Learning. [s.l.]: JMLR.org: 4904-4916, articleTitle=Scaling up visual and vision-language representation learning with noisy text supervision, refAbstract=null), Reference(id=1322961026500665536, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2023, volume=28, issue=5, pageStart=1384, pageEnd=1395, url=null, language=null, rfNumber=null, rfOrder=10, authorNames=Jiang D, Ye M, journalName=Journal of Image and Graphics, refType=null, unstructuredReference=Jiang D and Ye M. 2023. Transformer network for cross-modal text-to-image person re-identification. Journal of Image and Graphics,28(5): 1384-1395, articleTitle=Transformer network for cross-modal text-to-image person re-identification, refAbstract=null), Reference(id=1322961026571968705, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2023, volume=28, issue=5, pageStart=1384, pageEnd=1395, url=null, language=null, rfNumber=null, rfOrder=11, authorNames=姜定, 叶茫, journalName=中国图象图形学报, refType=null, unstructuredReference=姜定,叶茫. 2023. 面向跨模态文本到图像行人重识别的Transformer网络. 中国图象图形学报,28(5): 1384-1395 [DOI: 10.11834/jig.220620], articleTitle=面向跨模态文本到图像行人重识别的Transformer网络, refAbstract=null), Reference(id=1322961026651660482, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=5583, pageEnd=5594, url=null, language=null, rfNumber=null, rfOrder=12, authorNames=Kim W, Son B, Kim I, journalName=JMLR, refType=null, unstructuredReference=Kim W,Son B and Kim I. 2021. ViLT: vision-and-language transformer without convolution or region supervision//Proceedings of the 38th International Conference on Machine Learning. [s.l.]: JMLR.org: 5583-5594, articleTitle=ViLT: vision-and-language transformer without convolution or region supervision, refAbstract=null), Reference(id=1322961026714575043, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=212, pageEnd=228, url=null, language=null, rfNumber=null, rfOrder=13, authorNames=Lee K H, Chen X, Hua G, Hu H D, He X D, journalName=null, refType=null, unstructuredReference=Lee K H,Chen X,Hua G,Hu H D and He X D. 2018. Stacked cross attention for image-text matching//Proceedings of the 15th European Conference on Computer Vision. Munich,Germany: Springer: 212-228 [DOI: 10.1007/978-3-030-01225-0_13], articleTitle=Stacked cross attention for image-text matching, refAbstract=null), Reference(id=1322961028388102340, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=#814, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=14, authorNames=Li J N, Li D X, Savarese S, Hoi S, journalName=null, refType=null, unstructuredReference=Li J N,Li D X,Savarese S and Hoi S. 2023. BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models//Proceedings of the 40th International Conference on Machine Learning. Honolulu,USA: JMLR.org: #814, articleTitle=BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models, refAbstract=null), Reference(id=1322961028459405509, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2025, volume=null, issue=null, pageStart=2851, pageEnd=2859, url=null, language=null, rfNumber=null, rfOrder=15, authorNames=Liang Z Y, Liang M Y, Huang W, Li Y W, Liu W, Shao Y X, journalName=null, refType=null, unstructuredReference=Liang Z Y,Liang M Y,Huang W,Li Y W,Liu W,Shao Y X,et al. 2025. Dynamic self-adaptive multiscale distillation from pre-trained multimodal large model for efficient cross-modal retrieval//Proceedings of the 33rd ACM International Conference on Multimedia. Dublin,Ireland: ACM: 2851-2859 [DOI: 10.1145/3746027.3754844], articleTitle=Dynamic self-adaptive multiscale distillation from pre-trained multimodal large model for efficient cross-modal retrieval, refAbstract=null), Reference(id=1322961028551680198, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2015, volume=null, issue=null, pageStart=null, pageEnd=null, url=https://arxiv.org/pdf/1506.00019.pdf, language=null, rfNumber=null, rfOrder=16, authorNames=Lipton Z C, Berkowitz J, Elkan C, journalName=null, refType=null, unstructuredReference=Lipton Z C,Berkowitz J and Elkan C. 2015. A critical review of recurrent neural networks for sequence learning [EB/OL]. [2025-10-28]., articleTitle=A critical review of recurrent neural networks for sequence learning, refAbstract=null), Reference(id=1322961028631371975, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=null, pageEnd=null, url=https://arxiv.org/pdf/1907.11692.pdf, language=null, rfNumber=null, rfOrder=17, authorNames=Liu Y H, Ott M, Goyal N, Du J F, Joshi M, Chen D Q, journalName=null, refType=null, unstructuredReference=Liu Y H,Ott M,Goyal N,Du J F,Joshi M,Chen D Q,et al. 2019. RoBERTa: a robustly optimized BERT pretraining approach [EB/OL]. [2025-10-28]., articleTitle=RoBERTa: a robustly optimized BERT pretraining approach, refAbstract=null), Reference(id=1322961028711063752, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=18, authorNames=Loshchilov I, Hutter F, journalName=null, refType=null, unstructuredReference=Loshchilov I and Hutter F. 2019. Decoupled weight decay regularization//Proceedings of the 7th International Conference on Learning Representations. New Orleans,USA: ICLR, articleTitle=Decoupled weight decay regularization, refAbstract=null), Reference(id=1322961028790755529, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=23803, pageEnd=23828, url=null, language=null, rfNumber=null, rfOrder=19, authorNames=Mao A Q, Mohri M, Zhong Y T, journalName=null, refType=null, unstructuredReference=Mao A Q,Mohri M and Zhong Y T. 2023. Cross-entropy loss functions: theoretical analysis and applications//Proceedings of the 40th International Conference on Machine Learning. Honolulu,USA: JMLR.org: 23803-23828, articleTitle=Cross-entropy loss functions: theoretical analysis and applications, refAbstract=null), Reference(id=1322961028878835914, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=4948, pageEnd=4956, url=null, language=null, rfNumber=null, rfOrder=20, authorNames=Qin Y, Peng D Z, Peng X, Wang X, Hu P, journalName=null, refType=null, unstructuredReference=Qin Y,Peng D Z,Peng X,Wang X and Hu P. 2022. Deep evidential learning with noisy correspondence for cross-modal retrieval//Proceedings of the 30th ACM International Conference on Multimedia. Lisboa,Portugal: ACM: 4948-4956 [DOI: 10.1145/3503161.3547922], articleTitle=Deep evidential learning with noisy correspondence for cross-modal retrieval, refAbstract=null), Reference(id=1322961028962721995, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2009, volume=null, issue=7, pageStart=#311, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=21, authorNames=Qu C, Zhai J X, journalName=null, refType=null, unstructuredReference=Qu C and Zhai J X. 2009. Analysis of Han dynasty stone reliefs in Xuchang. Legal System and Society,(7): #311, articleTitle=Analysis of Han dynasty stone reliefs in Xuchang, refAbstract=null), Reference(id=1322961029034025164, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2009, volume=7, issue=null, pageStart=null, pageEnd=311, url=null, language=null, rfNumber=null, rfOrder=22, authorNames=渠超, 翟京襄, journalName=法制与社会, refType=null, unstructuredReference=渠超,翟京襄. 2009. 浅析许昌汉画像石. 法制与社会,7: #311 [DOI: 10.19387/j.cnki.1009-0592.2009.07.201], articleTitle=浅析许昌汉画像石, refAbstract=null), Reference(id=1322961029134688461, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=8748, pageEnd=8763, url=null, language=null, rfNumber=null, rfOrder=23, authorNames=Radford A, Kim J W, Hallacy C, Ramesh A, Goh G, Agarwal S, journalName=JMLR, refType=null, unstructuredReference=Radford A,Kim J W,Hallacy C,Ramesh A,Goh G,Agarwal S,et al. 2021. Learning transferable visual models from natural language supervision//Proceedings of the 38th International Conference on Machine Learning. [s.l.]: JMLR.org: 8748-8763, articleTitle=Learning transferable visual models from natural language supervision, refAbstract=null), Reference(id=1322961029226963150, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2015, volume=null, issue=null, pageStart=91, pageEnd=99, url=null, language=null, rfNumber=null, rfOrder=24, authorNames=Ren S Q, He K M, Girshick R, Sun J, journalName=null, refType=null, unstructuredReference=Ren S Q,He K M,Girshick R and Sun J. 2015. Faster R-CNN: towards real-time object detection with region proposal networks//Proceedings of the 29th International Conference on Neural Information Processing Systems. Montreal,Canada: MIT Press: 91-99, articleTitle=Faster R-CNN: towards real-time object detection with region proposal networks, refAbstract=null), Reference(id=1322961029294072015, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2000, volume=22, issue=12, pageStart=1349, pageEnd=1380, url=null, language=null, rfNumber=null, rfOrder=25, authorNames=Smeulders A W M, Worring M, Santini S, Gupta A, Jain R, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Smeulders A W M,Worring M,Santini S,Gupta A and Jain R. 2000. Content-based image retrieval at the end of the early years. IEEE Transactions on Pattern Analysis and Machine Intelligence,22(12): 1349-1380 [DOI: 10.1109/34.895972], articleTitle=Content-based image retrieval at the end of the early years, refAbstract=null), Reference(id=1322961029361180880, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2025, volume=30, issue=12, pageStart=3707, pageEnd=3739, url=null, language=null, rfNumber=null, rfOrder=26, authorNames=Song W T, Liao L Y, Zhang H T, Li L, Yu T X, Zhao Y S, journalName=Journal of Image and Graphics, refType=null, unstructuredReference=Song W T,Liao L Y,Zhang H T,Li L,Yu T X,Zhao Y S,et al. 2025. Applications and prospects of artificial intelligence in the cultural heritage. Journal of Image and Graphics,30(12): 3707-3739, articleTitle=Applications and prospects of artificial intelligence in the cultural heritage, refAbstract=null), Reference(id=1322961029424095441, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=null, volume=30, issue=12, pageStart=3707, pageEnd=3739, url=null, language=null, rfNumber=null, rfOrder=27, authorNames=宋维涛, 廖聆宇, 张浩天, 李琳, 俞天秀, 赵永生, journalName=人工智能在文物行业的应用与展望. 中国图象图形学报, refType=null, unstructuredReference=宋维涛,廖聆宇,张浩天,李琳,俞天秀,赵永生,等. 2025. 人工智能在文物行业的应用与展望. 中国图象图形学报,30(12): 3707-3739 [DOI: 10.11834/jig.240765], articleTitle=2025, refAbstract=null), Reference(id=1322961029499592914, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=null, pageEnd=null, url=https://arxiv.org/pdf/1807.03748.pdf, language=null, rfNumber=null, rfOrder=28, authorNames=van den Oord A, Li Y Z, Vinyals O, journalName=null, refType=null, unstructuredReference=van den Oord A,Li Y Z and Vinyals O. 2019. Representation learning with contrastive predictive coding [EB/OL]. [2025-10-28]., articleTitle=Representation learning with contrastive predictive coding, refAbstract=null), Reference(id=1322961029600256211, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=6000, pageEnd=6010, url=null, language=null, rfNumber=null, rfOrder=29, authorNames=Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A N, journalName=null, refType=null, unstructuredReference=Vaswani A,Shazeer N,Parmar N,Uszkoreit J,Jones L,Gomez A N,et al. 2017. Attention is all you need//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach,USA: Curran Associates Inc.: 6000-6010, articleTitle=Attention is all you need, refAbstract=null), Reference(id=1322961029742862548, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=#412, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=30, authorNames=Wang J K, Chen D D, Wu Z X, Luo C, Zhou L W, Zhao Y C, journalName=null, refType=null, unstructuredReference=Wang J K,Chen D D,Wu Z X,Luo C,Zhou L W,Zhao Y C,et al. 2022. OmniVL: one foundation model for image-language and video-language tasks//Proceedings of the 36th International Conference on Neural Information Processing Systems. New Orleans,USA: Curran Associates Inc.: #412, articleTitle=OmniVL: one foundation model for image-language and video-language tasks, refAbstract=null), Reference(id=1322961029822554325, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=19175, pageEnd=19186, url=null, language=null, rfNumber=null, rfOrder=31, authorNames=Wang W H, Bao H B, Dong L, Bjorck J, Peng Z L, Liu Q, journalName=null, refType=null, unstructuredReference=Wang W H,Bao H B,Dong L,Bjorck J,Peng Z L,Liu Q,et al. 2023. Image as a foreign language: BEIT pretraining for vision and vision-language tasks//Proceedings of 2023 Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver,Canada: IEEE: 19175-19186 [DOI: 10.1109/CVPR52729.2023.01838], articleTitle=Image as a foreign language: BEIT pretraining for vision and vision-language tasks, refAbstract=null), Reference(id=1322961029914829014, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=null, pageEnd=null, url=https://arxiv.org/pdf/2211.01335.pdf, language=null, rfNumber=null, rfOrder=32, authorNames=Yang A, Pan J S, Lin J Y, Men R, Zhang Y C, Zhou J R, journalName=null, refType=null, unstructuredReference=Yang A,Pan J S,Lin J Y,Men R,Zhang Y C,Zhou J R,et al. 2023. Chinese CLIP: contrastive vision-language pretraining in Chinese [EB/OL]. [2025-10-28]., articleTitle=Chinese CLIP: contrastive vision-language pretraining in Chinese, refAbstract=null), Reference(id=1322961030225207511, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2021, volume=26, issue=6, pageStart=1368, pageEnd=1388, url=null, language=null, rfNumber=null, rfOrder=33, authorNames=Yin Q Y, Huang Y, Zhang J G, Wu S, Wang L, journalName=Journal of Image and Graphics, refType=null, unstructuredReference=Yin Q Y,Huang Y,Zhang J G,Wu S and Wang L. 2021. Survey on deep learning based cross-modal retrieval. Journal of Image and Graphics,26(6): 1368-1388, articleTitle=Survey on deep learning based cross-modal retrieval, refAbstract=null), Reference(id=1322961030292316376, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2021, volume=26, issue=6, pageStart=1368, pageEnd=1388, url=null, language=null, rfNumber=null, rfOrder=34, authorNames=尹奇跃, 黄岩, 张俊格, 吴书, 王亮, journalName=中国图象图形学报, refType=null, unstructuredReference=尹奇跃,黄岩,张俊格,吴书,王亮. 2021. 基于深度学习的跨模态检索综述. 中国图象图形学报,26(6): 1368-1388 [DOI: 10.11834/jig.200862], articleTitle=基于深度学习的跨模态检索综述, refAbstract=null), Reference(id=1322961030392979673, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, doi=null, pmid=null, pmcid=null, year=2024, volume=46, issue=5, pageStart=3156, pageEnd=3168, url=null, language=null, rfNumber=null, rfOrder=35, authorNames=Zeng Y, Zhang X S, Li H, Wang J W, Zhang J P, Zhou W C S, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Zeng Y,Zhang X S,Li H,Wang J W,Zhang J P and Zhou W C S. 2024. X2-VLM: all-in-one pre-trained model for vision-language tasks. IEEE Transactions on Pattern Analysis and Machine Intelligence,46(5): 3156-3168 [DOI: 10.1109/TPAMI.2023.3339661], articleTitle=X2-VLM: all-in-one pre-trained model for vision-language tasks, refAbstract=null)], funds=[Fund(id=1322961025481449652, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, awardId=null, language=EN, fundingSource=National Natural Science Foundation of China(62162065);Joint Special Project Research Foundation of Yunnan Province(202401BF070001-023);Yunnan Province Visual and Cultural Innovation Team(202505AS350009);Yunnan Fundamental Research Projects(202201AT070167), fundOrder=null, country=null), Fund(id=1322961025569530037, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, awardId=null, language=CN, fundingSource=国家自然科学基金项目(62162065);云南大学“双一流”建设联合专项(202401BF070001-023);云南省“视觉与文化计算创新团队”项目(202505AS350009);云南省科技厅应用基础研究计划项目(202201AT070167), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1322961017294168182, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, xref=null, ext=[AuthorCompanyExt(id=1322961017302556791, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Information Science and Engineering,Yunnan University,Kunming 650504,China), AuthorCompanyExt(id=1322961017310945400, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=云南大学信息学院,昆明 650504)])], figs=[ArticleFig(id=1322961020666388626, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.1, caption=Examples of Han portrait stones, figureFileSmall=IzR8mfx8P8EvQ7OF874kGA==, figureFileBig=lYnp/pVqgXO1dC7xP8MsTg==, tableContent=null), ArticleFig(id=1322961020746080403, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图1, caption=汉画像石实例, figureFileSmall=IzR8mfx8P8EvQ7OF874kGA==, figureFileBig=lYnp/pVqgXO1dC7xP8MsTg==, tableContent=null), ArticleFig(id=1322961021110984852, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.2, caption=Overall structure diagram of the model, figureFileSmall=FqVQTIbxUYbSoxbLtsR4ZA==, figureFileBig=Nq4fagBS+UUG/D0rmEAArw==, tableContent=null), ArticleFig(id=1322961021354254485, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图2, caption=模型总体结构图, figureFileSmall=FqVQTIbxUYbSoxbLtsR4ZA==, figureFileBig=Nq4fagBS+UUG/D0rmEAArw==, tableContent=null), ArticleFig(id=1322961021429751958, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.3, caption=Structure diagram of the image encoder, figureFileSmall=XcpBLsXE9tGtmfB20orqrQ==, figureFileBig=Q2z2jZ4A9KVQCHGgyOpBXw==, tableContent=null), ArticleFig(id=1322961021496860823, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图3, caption=图像编码器结构图, figureFileSmall=XcpBLsXE9tGtmfB20orqrQ==, figureFileBig=Q2z2jZ4A9KVQCHGgyOpBXw==, tableContent=null), ArticleFig(id=1322961021559775384, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.4, caption=Picture of chariots and horses on the move, figureFileSmall=39KfU4SRfbXe8F+pGcS5iA==, figureFileBig=evv+6ESjVHQPI2sFRmJHJw==, tableContent=null), ArticleFig(id=1322961021656244377, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图4, caption=车马出行图, figureFileSmall=39KfU4SRfbXe8F+pGcS5iA==, figureFileBig=evv+6ESjVHQPI2sFRmJHJw==, tableContent=null), ArticleFig(id=1322961021765296282, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.5, caption=Structure diagram of the text encoder, figureFileSmall=q8jGGh1ZikiYxHW2p9UdWQ==, figureFileBig=YzDYUObuE+kp4r+ugEpEdg==, tableContent=null), ArticleFig(id=1322961021840793755, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图5, caption=文本编码器结构图, figureFileSmall=q8jGGh1ZikiYxHW2p9UdWQ==, figureFileBig=YzDYUObuE+kp4r+ugEpEdg==, tableContent=null), ArticleFig(id=1322961021903708316, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.6, caption=Structure diagram of the semantic enhancement module, figureFileSmall=TMSDdQTylD3K9nv3cGPTzw==, figureFileBig=XSNcAvLlR27CYYRGpDMWBw==, tableContent=null), ArticleFig(id=1322961021987594397, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图6, caption=语义增强模块结构图, figureFileSmall=TMSDdQTylD3K9nv3cGPTzw==, figureFileBig=XSNcAvLlR27CYYRGpDMWBw==, tableContent=null), ArticleFig(id=1322961022063091870, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.7, caption=Structure diagram of semantic embedding random template text reorganization, figureFileSmall=oCh8JCbsiJJv3Q485p+T/A==, figureFileBig=F5jIqsgpP3+AbWTz/QEERw==, tableContent=null), ArticleFig(id=1322961022159560863, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图7, caption=语义嵌入随机模板文本重组结构图, figureFileSmall=oCh8JCbsiJJv3Q485p+T/A==, figureFileBig=F5jIqsgpP3+AbWTz/QEERw==, tableContent=null), ArticleFig(id=1322961022226669728, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.8, caption=Examples of Han portrait stones dataset processing, figureFileSmall=vHIoiLanLlXPnOT/KM84Yw==, figureFileBig=VFL45ahM4xSqcPIi/qwkyQ==, tableContent=null), ArticleFig(id=1322961022297972897, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图8, caption=汉画像石数据集处理示例

((a)origin image;(b)after cutting;(c)after removing the blur;(d)after clarification)

, figureFileSmall=vHIoiLanLlXPnOT/KM84Yw==, figureFileBig=VFL45ahM4xSqcPIi/qwkyQ==, tableContent=null), ArticleFig(id=1322961023963111586, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.9, caption=Example diagram of object extraction for Han portrait stones, figureFileSmall=nb5IAljsG4N5D/ZRcWQ7eg==, figureFileBig=ZTsV7CCn9QzQJSxAQv7+Ug==, tableContent=null), ArticleFig(id=1322961024034414755, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图9, caption=汉画像石对象提取示例图, figureFileSmall=nb5IAljsG4N5D/ZRcWQ7eg==, figureFileBig=ZTsV7CCn9QzQJSxAQv7+Ug==, tableContent=null), ArticleFig(id=1322961024114106532, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.10, caption=Examples of image-text retrieval dataset for Han portrait stones, figureFileSmall=KJ9k2W0T7XBbDV4yJrw1FQ==, figureFileBig=h6k6QBhHY8st2wCKp9BvyQ==, tableContent=null), ArticleFig(id=1322961024181215397, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图10, caption=汉画像石图文检索数据集示例, figureFileSmall=KJ9k2W0T7XBbDV4yJrw1FQ==, figureFileBig=h6k6QBhHY8st2wCKp9BvyQ==, tableContent=null), ArticleFig(id=1322961024281878694, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.11, caption=Visualization comparison of positive and negative sample separation, figureFileSmall=hT7Nvh4rjjeN/rkMIPY/Sg==, figureFileBig=0grLpcgLVyfJNuqKzoH53g==, tableContent=null), ArticleFig(id=1322961024344793255, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图11, caption=正负样本分离度可视化对比图

((a)initial model;(b)semantic enhancement model;(c)text reorganization model;(d)semantic enhancement and text reorganization)

, figureFileSmall=hT7Nvh4rjjeN/rkMIPY/Sg==, figureFileBig=0grLpcgLVyfJNuqKzoH53g==, tableContent=null), ArticleFig(id=1322961024445456552, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.12, caption=Visualization experiment comparison diagram ((a) text2img retrieval; (b) img2text retrieval), figureFileSmall=uTaCSoly83pi/5QfcZesgQ==, figureFileBig=XeK0fL7q7Z2IR/cUZg+LKg==, tableContent=null), ArticleFig(id=1322961024516759721, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图12, caption=可视化实验对比图, figureFileSmall=uTaCSoly83pi/5QfcZesgQ==, figureFileBig=XeK0fL7q7Z2IR/cUZg+LKg==, tableContent=null), ArticleFig(id=1322961024592257194, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Fig.13, caption=Visualization experiment diagram of data robustness((a)img2text retrieval;(b)text2img retrieval), figureFileSmall=DrICj2+FoXh+6lPD+PUiyw==, figureFileBig=BBgTYGDxwmz4SjMGr9irXw==, tableContent=null), ArticleFig(id=1322961024667754667, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=图13, caption=鲁棒性可视化实验图, figureFileSmall=DrICj2+FoXh+6lPD+PUiyw==, figureFileBig=BBgTYGDxwmz4SjMGr9irXw==, tableContent=null), ArticleFig(id=1322961024739057836, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Table 1, caption=Comparison of indicators under fusion of different feature proportions, figureFileSmall=null, figureFileBig=null, tableContent=
特征比例文本到图像检索图像到文本检索
原始特征注意力特征增强特征r@1/%r@5/%r@10/%r@1/%r@5/%r@10/%
0.40.10.545.1981.7392.3046.4286.3094.64
0.40.20.449.0381.7392.3044.6480.9595.83
0.40.30.348.0781.7395.1939.2884.5292.26
0.40.40.248.0785.5795.1947.1284.5293.45
0.50.10.448.0785.5795.1952.3884.5293.45
0.50.20.349.0385.5795.1945.2383.3394.64
0.50.30.246.1579.8088.4635.7175.5991.07
0.50.40.142.3081.7391.3441.6679.1689.88
0.60.10.347.1177.8889.4247.0279.1695.23
0.60.20.244.2375.9688.4639.2881.5494.04
0.60.30.143.2679.8092.3042.2685.1194.64
0.70.10.249.0381.7394.2342.8584.5294.04
0.70.20.144.2376.9290.3836.3073.2189.28
0.80.10.140.3878.8489.4243.4579.1691.07
0.40.50.155.7687.5795.1952.3885.7197.02
), ArticleFig(id=1322961024831332525, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=表1, caption=

不同特征比例融合下的指标对比

, figureFileSmall=null, figureFileBig=null, tableContent=
特征比例文本到图像检索图像到文本检索
原始特征注意力特征增强特征r@1/%r@5/%r@10/%r@1/%r@5/%r@10/%
0.40.10.545.1981.7392.3046.4286.3094.64
0.40.20.449.0381.7392.3044.6480.9595.83
0.40.30.348.0781.7395.1939.2884.5292.26
0.40.40.248.0785.5795.1947.1284.5293.45
0.50.10.448.0785.5795.1952.3884.5293.45
0.50.20.349.0385.5795.1945.2383.3394.64
0.50.30.246.1579.8088.4635.7175.5991.07
0.50.40.142.3081.7391.3441.6679.1689.88
0.60.10.347.1177.8889.4247.0279.1695.23
0.60.20.244.2375.9688.4639.2881.5494.04
0.60.30.143.2679.8092.3042.2685.1194.64
0.70.10.249.0381.7394.2342.8584.5294.04
0.70.20.144.2376.9290.3836.3073.2189.28
0.80.10.140.3878.8489.4243.4579.1691.07
0.40.50.155.7687.5795.1952.3885.7197.02
), ArticleFig(id=1322961024906829998, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Table 2, caption=Comparison of metrics between the method in this paper and other methods, figureFileSmall=null, figureFileBig=null, tableContent=
方法文本到图像检索图像到文本检索
Mean_rr@1r@5r@10Mean_rr@1r@5r@10
VSE(Faghri等,2018)38.1322.1142.3050.0038.6819.6441.6654.76
ViLT(Kim等,2021)39.6711.9045.2361.9038.7710.5745.1960.57
X2-VLM(Zeng等,2024)42.9416.3447.1165.3860.9029.1667.2686.30
BEiT3(Wang等,2023)59.6128.8467.3082.6959.2926.9264.4286.53
BLIP(Li等,2023)64.2833.9273.2185.7148.3921.1555.7668.26
Chinese-CLIP(Yang等,2023)46.7921.1549.0370.1935.7416.0737.5053.57
DSMD(Liang等,2025)34.5210.7139.8852.9733.5213.0936.9050.59
3SHNet(Ge等,2024)55.1227.8861.5375.9660.3129.1668.4583.33
本文79.4855.7687.5095.1978.3752.3885.7197.02
), ArticleFig(id=1322961024994910383, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=表2, caption=

本文方法与其他方法指标对比

, figureFileSmall=null, figureFileBig=null, tableContent=
方法文本到图像检索图像到文本检索
Mean_rr@1r@5r@10Mean_rr@1r@5r@10
VSE(Faghri等,2018)38.1322.1142.3050.0038.6819.6441.6654.76
ViLT(Kim等,2021)39.6711.9045.2361.9038.7710.5745.1960.57
X2-VLM(Zeng等,2024)42.9416.3447.1165.3860.9029.1667.2686.30
BEiT3(Wang等,2023)59.6128.8467.3082.6959.2926.9264.4286.53
BLIP(Li等,2023)64.2833.9273.2185.7148.3921.1555.7668.26
Chinese-CLIP(Yang等,2023)46.7921.1549.0370.1935.7416.0737.5053.57
DSMD(Liang等,2025)34.5210.7139.8852.9733.5213.0936.9050.59
3SHNet(Ge等,2024)55.1227.8861.5375.9660.3129.1668.4583.33
本文79.4855.7687.5095.1978.3752.3885.7197.02
), ArticleFig(id=1322961025082990768, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Table 3, caption=Ablation experiment, figureFileSmall=null, figureFileBig=null, tableContent=
模块文本到图像检索图像到文本检索
多阶段特征融合文本重组动态温度r@1r@5r@10r@1r@5r@10
×××14.4240.3852.8811.3030.3547.02
××√24.0344.2356.7312.5034.5252.38
×√×37.5070.1983.6332.7368.4591.07
×√√36.5372.1189.4238.0974.4089.28
√××29.8060.5773.0721.4257.1474.40
√×√38.4662.5075.9623.2150.0067.85
√√×42.3077.8890.3842.2680.3594.04
√√√55.7687.5095.1952.3885.7197.02
), ArticleFig(id=1322961025175265457, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=表3, caption=

消融实验

, figureFileSmall=null, figureFileBig=null, tableContent=
模块文本到图像检索图像到文本检索
多阶段特征融合文本重组动态温度r@1r@5r@10r@1r@5r@10
×××14.4240.3852.8811.3030.3547.02
××√24.0344.2356.7312.5034.5252.38
×√×37.5070.1983.6332.7368.4591.07
×√√36.5372.1189.4238.0974.4089.28
√××29.8060.5773.0721.4257.1474.40
√×√38.4662.5075.9623.2150.0067.85
√√×42.3077.8890.3842.2680.3594.04
√√√55.7687.5095.1952.3885.7197.02
), ArticleFig(id=1322961025259151538, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=EN, label=Table 4, caption=Comparison of data robustness experiments, figureFileSmall=null, figureFileBig=null, tableContent=
方法文本到图像检索图像到文本检索
Mean_rr@1r@5r@10Mean_rr@1r@5r@10
图像衰减72.1142.3082.6991.3473.0141.6682.7394.64
文本衰减75.9650.9686.5390.3877.9751.1987.5095.23
图像和文本衰减70.1942.3076.9291.3474.0042.8585.7193.45
本文79.4855.7687.5095.1978.3752.3885.7197.02
), ArticleFig(id=1322961025355620531, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, language=CN, label=表4, caption=

鲁棒性实验对比

, figureFileSmall=null, figureFileBig=null, tableContent=
方法文本到图像检索图像到文本检索
Mean_rr@1r@5r@10Mean_rr@1r@5r@10
图像衰减72.1142.3082.6991.3473.0141.6682.7394.64
文本衰减75.9650.9686.5390.3877.9751.1987.5095.23
图像和文本衰减70.1942.3076.9291.3474.0042.8585.7193.45
本文79.4855.7687.5095.1978.3752.3885.7197.02
)], attaches=null, journal=Journal(id=1249023527618129992, delFlag=0, nameCn=中国图象图形学报, nameEn=Journal of Image and Graphics, nameHistory1=null, nameHistory2=null, issn=1006-8961, eissn=null, cn=11-3758, coden=CODEN ZTTXFZ, periodic=0, language=CN, oaType=null, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=uirXtX858YS3zEpFXZttJA==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Image and Graphics, journalRemark=null, publicationField=null, createdTime=1775720014721, updatedTime=1788951488513, createdBy=18614031015, updatedBy=13041195026, firstLetterCn=Z, firstLetterEn=Z, subjectCode=Electronic Information, subjectName=null, subjectCodeEn=Electronic Information, subjectNameEn=null, picCn=uirXtX858YS3zEpFXZttJA==, picEn=bud7qaxfvWHeFsbyBTAiKQ==, jcr=null, cjcr=null, exts=[JournalExt(id=1304520351199814396, language=CN, name=中国图象图形学报, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1788951488542, updatedTime=1788951488542, createdBy=13041195026, updatedBy=13041195026, submissionGuidelinesUrl=, submissionAuthorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-author&redirect_uri=https%3A%2F%2Fcjig.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=e6369def-2842-41d8, submissionEditorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-editor&redirect_uri=https%3A%2F%2Fcjigeditor.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=9ccec05b-6bd, submissionReviewUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-author&redirect_uri=https%3A%2F%2Fcjig.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=1e8a31c8-5434-4f78, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1304520351250146045, language=EN, name=Journal of Image and Graphics, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1788951488554, updatedTime=1788951488554, createdBy=13041195026, updatedBy=13041195026, submissionGuidelinesUrl=, submissionAuthorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-author&redirect_uri=https%3A%2F%2Fcjig.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=e6369def-2842-41d8, submissionEditorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-editor&redirect_uri=https%3A%2F%2Fcjigeditor.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=9ccec05b-6bd, submissionReviewUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-author&redirect_uri=https%3A%2F%2Fcjig.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=1e8a31c8-5434-4f78, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1249024232475115590, websiteList=[Website(id=1249025782459334881, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1249024232475115590, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN, language=CN, createTime=1775720552315, createBy=18614031015, updateTime=1775720586268, updateBy=18614031015, name=中国图象图形学报-中文, tplId=1146099689490845704, title=中国图象图形学报, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1249026166254928133, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=articleTextType, value=kx, createTime=1775720643819, updateTime=1775720643819, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166221373698, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=banner, value=null, createTime=1775720643811, updateTime=1775720643811, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166271705352, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=grayFlag, value=0, createTime=1775720643823, updateTime=1775720643823, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166212985089, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=logo, value=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/file/pic?fileId=TDRjKTHfgAnvFKZaDA70wA==, createTime=1775720643809, updateTime=1775720643809, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166288482570, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=minRunFlag, value=0, createTime=1775720643827, updateTime=1775720643827, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166246539524, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/file/pic, createTime=1775720643817, updateTime=1775720643817, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166280093961, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=silenceFlag, value=0, createTime=1775720643825, updateTime=1775720643825, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166233956611, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1775720643814, updateTime=1775720643814, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166259122438, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=themeColor, value=null, createTime=1775720643820, updateTime=1775720643820, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166267511047, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=themeStyle, value=null, createTime=1775720643822, updateTime=1775720643822, creator=18614031015, updator=18614031015)]), Website(id=1249025782681633001, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1249024232475115590, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN, language=EN, createTime=1775720552368, createBy=18614031015, updateTime=1775720607118, updateBy=18614031015, name=中国图象图形学报-英文, tplId=1146101810881728533, title=Journal of Image and Graphics, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1249026195371786511, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=articleTextType, value=kx, createTime=1775720650761, updateTime=1775720650761, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195355009292, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=banner, value=null, createTime=1775720650757, updateTime=1775720650757, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195392758034, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=grayFlag, value=0, createTime=1775720650766, updateTime=1775720650766, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195342426379, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=logo, value=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN/file/pic?fileId=TDRjKTHfgAnvFKZaDA70wA==, createTime=1775720650754, updateTime=1775720650754, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195409535252, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=minRunFlag, value=0, createTime=1775720650770, updateTime=1775720650770, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195367592206, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN/file/pic, createTime=1775720650760, updateTime=1775720650760, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195401146643, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=silenceFlag, value=0, createTime=1775720650768, updateTime=1775720650768, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195359203597, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1775720650758, updateTime=1775720650758, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195380175120, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=themeColor, value=null, createTime=1775720650763, updateTime=1775720650763, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195388563729, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=themeStyle, value=null, createTime=1775720650765, updateTime=1775720650765, creator=18614031015, updator=18614031015)])], journalTitle=中国图象图形学报, weixinUrl=null, journalUrl=https://www.cjig.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Image and Graphics, journalPhotoCn=uirXtX858YS3zEpFXZttJA==, journalPhotoEn=bud7qaxfvWHeFsbyBTAiKQ==, journalFirstLetter=Z, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=, provinceCode=null, provinceName=null, collectFlag=false, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/10.11834/jig.250393, detailUrlEn=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN/10.11834/jig.250393, pdfUrlCn=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/PDF/10.11834/jig.250393, pdfUrlEn=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN/PDF/10.11834/jig.250393, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1786809600000, fullTextJson=null, articleText=null, reference=null)
收藏切换
语义增强与文本重组协同的汉画像石图文检索
收藏切换
PDF下载
姜坤 , 钱文华 * , 刘朋
中国图象图形学报 | 图像理解和计算机视觉 2026,31(8): 3030-3044
收起
收藏切换
中国图象图形学报 |图像理解和计算机视觉 2026 , 31 (8) : 3030 -3044
语义增强与文本重组协同的汉画像石图文检索
全屏
[Author(id=1322961017378054266, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=2476270892@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322961017453551740, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961017378054266, language=EN, stringName=Kun Jiang, firstName=Kun, middleName=null, lastName=Jiang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Information Science and Engineering,Yunnan University,Kunming 650504,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322961017533243517, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961017378054266, language=CN, stringName=姜坤, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=云南大学信息学院,昆明 650504, bio={"content":"

姜坤,男,硕士研究生,主要研究方向为深度学习和计算机视觉。E-mail:

"}, bioImg=null, bioContent=

姜坤,男,硕士研究生,主要研究方向为深度学习和计算机视觉。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322961017294168182, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, xref=null, ext=[AuthorCompanyExt(id=1322961017302556791, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Information Science and Engineering,Yunnan University,Kunming 650504,China), AuthorCompanyExt(id=1322961017310945400, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=云南大学信息学院,昆明 650504)])]), Author(id=1322961017612935295, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=whqian@ynu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1322961017730375809, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961017612935295, language=EN, stringName=Wenhua Qian, firstName=Wenhua, middleName=null, lastName=Qian, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=School of Information Science and Engineering,Yunnan University,Kunming 650504,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322961017784901762, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961017612935295, language=CN, stringName=钱文华, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=云南大学信息学院,昆明 650504, bio={"content":"

钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:

"}, bioImg=null, bioContent=

钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322961017294168182, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, xref=null, ext=[AuthorCompanyExt(id=1322961017302556791, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Information Science and Engineering,Yunnan University,Kunming 650504,China), AuthorCompanyExt(id=1322961017310945400, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=云南大学信息学院,昆明 650504)])]), Author(id=1322961019466817668, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=pengliu0606@gmail.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322961019563286662, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961019466817668, language=EN, stringName=Peng Liu, firstName=Peng, middleName=null, lastName=Liu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=School of Information Science and Engineering,Yunnan University,Kunming 650504,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322961019638784135, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, authorId=1322961019466817668, language=CN, stringName=刘朋, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=云南大学信息学院,昆明 650504, bio={"content":"

刘朋,男,博士研究生,主要研究方向为计算机视觉和图像处理。E-mail:

"}, bioImg=null, bioContent=

刘朋,男,博士研究生,主要研究方向为计算机视觉和图像处理。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322961017294168182, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, xref=null, ext=[AuthorCompanyExt(id=1322961017302556791, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=School of Information Science and Engineering,Yunnan University,Kunming 650504,China), AuthorCompanyExt(id=1322961017310945400, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1322961012428775513, companyId=1322961017294168182, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=云南大学信息学院,昆明 650504)])])]
姜坤 , 钱文华* , 刘朋
作者信息
  • 云南大学信息学院,昆明 650504
通讯作者:
钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:
作者简介:

姜坤,男,硕士研究生,主要研究方向为深度学习和计算机视觉。E-mail:

钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:

刘朋,男,博士研究生,主要研究方向为计算机视觉和图像处理。E-mail:

Image-text retrieval for Han portrait stones through the synergy of semantic enhancement and text reorganization
Kun Jiang , Wenhua Qian* , Peng Liu
Affiliations
  • School of Information Science and Engineering,Yunnan University,Kunming 650504,China
出版时间: 2026-08-16 doi: 10.11834/jig.250393
文章导航
收藏切换
目的

汉画像石是解读汉代历史文化的珍贵载体。图文检索技术能推动汉代文化研究向数字化转型,为文化保护提供智能化解决方案。然而,现有图文检索方法应用于汉画像石时,存在领域偏移、图像与文本难以对齐等问题。本文提出一种语义增强与文本重组协同的图文检索方法以提高检索的准确性和全面性。

方法

本文通过编码器分别完成汉画像石图像与文本特征的提取。在图像处理阶段,基于汉画像石图像对应掩膜图像,计算各语义对象权重,进而推导得出注意力权重。通过残差连接获取注意力特征,并将所得注意力权重直接作为语义增强特征。通过多阶段特征融合将原始特征、注意力特征和增强特征融合得到组合特征以丰富特征表示。在文本处理阶段,将文本与提取的图像语义对象嵌入随机提示模板,生成结构更清晰的增强文本,增强文本的语义表达能力。同时,为适配本文数据集结构,提出动态温度调整以及非对称相似度计算的方法,基于特征相似度自适应调整温度。对图像到文本和文本到图像双向的相似度矩阵计算采用不同温度系数以提升跨模态匹配的准确性与稳定性。

结果

通过本文所提方法验证,文本到图像的检索指标Mean_Recall、Recall@1、Recall@5和Recall@10相较于排名第2的BLIP(bootstrapping language-image pre-training)模型分别提高了15.20%、21.84%、14.29%和9.48%。图像到文本的检索指标Mean_Recall、Recall@1、Recall@5和Recall@10相较于排名第2的X2-VLM(all-in-one pre-trained model for vision-language tasks)模型分别提高了17.47%、23.22%、18.45%和10.72%,汉画像石图文可视化检索也得到了更好的效果。

结论

本文方法显著增强了模型对汉画像石图像与其描述文本的对齐能力,有效提升了图文检索在准确性、全面性等方面的表现。

汉画像石  /  图文检索  /  语义增强  /  文本重组  /  动态温度
Objective

Han portrait stones are decorative stone carvings on funerary architectural structures,such as tombs and ancestral halls,dating back to the Han Dynasty. The contents of these stones depict mythological legends,institutional systems,and everyday life scenes,including chariot processions,kitchen banquets,textile farming,and other scenes that directly reflect the social practices of the Han period. Iconic figures,such as the Queen Mother of the West (Xiwangmu),Fuxi and Nüwa,and feathered immortals (Yuren),represent the Han people’s cosmology of “immortal souls”. In particular,Xiwangmu images are prevalent in the tombs of the middle and lower classes but seldom appear in noble tombs,reflecting hierarchical differences in beliefs. Themes,such as ancient sages and war scenes,which combine architectural practicality and artistic decoration,serve educational and commemorative functions,embodying realism and romanticism. This study proposes an image-text retrieval method that leverages semantic information from mask images of Han portrait stones to enhance images and texts,addressing the domain shift problem present in existing image-text retrieval methods. The method introduces a semantic enhancement module and integrates semantic information with randomly generated prompt templates for text reorganization. This design enables the model to learn additional discriminative features,understand different objects in images,align text with images,and improve the accuracy and comprehensiveness of image-text retrieval.

Method

This study utilizes the architecture of Chinese contrastive language-image pre-training (Chinese CLIP) using a vision transformer (ViT) as the image encoder and robustly optimized bidirectional encoder representations from Transformers (RoBERTa) model as the text encoder to extract image and text features. Semantic information from mask images and original text is used to calculate object-specific weights. Attention weights are derived by combining these weights with features from mask and original images. After a residual connection,these attention weights are used to generate attention features,while the weights themselves serve as semantic enhancement features. The original image features,attention features,and enhancement features are combined through a multi-stage feature fusion strategy at a ratio of 0.4:0.5:0.1 to form composite features. This design enables the model to focus on the primary subjects and important objects in images while enriching feature representation without information loss. Effective semantic objects are extracted from the mask images,and the original texts are embedded into randomly generated prompt templates incorporating these objects to produce enhanced texts. This process structures text descriptions,strengthens semantic expression,and improves matching with image features. Furthermore,dynamic temperature adjustment and asymmetric similarity calculation are introduced: the contrastive learning temperature is adaptively adjusted based on feature similarity (low temperatures applied to high-similarity samples to enhance discriminative learning and high temperatures applied to low-similarity samples to increase tolerance). The different temperature coefficients (0.9 for image-to-text and 1.1 for text-to-image) balance the matching difficulty in both directions,accounting for the distribution differences between image and text features to improve the accuracy and stability of cross-modal matching.

Result

In this study,a series of comparative experiments,ablation experiments,and visual comparison experiments is conducted on the proposed model to evaluate the model’s performance and the contributions of each module. Results show that,on the Han portrait stone test set,the proposed method significantly improves the text-to-image retrieval metrics Mean_Recall (%),Recall@1 (%),Recall@5 (%),and Recall@10 (%) by 15.20%,21.84%,14.29%,and 9.48%,respectively,compared with the second-ranked bootstrapping language-image pre-training model in terms of precision and comprehensiveness. During the image-to-text retrieval,the metrics Mean_Recall (%),Recall@1 (%),Recall@5 (%),and Recall@10 (%) are improved by 17.47%,23.22%,18.45%,and 10.72%,respectively,compared with the second-ranked cross-modal vision-language model (X2-VLM),demonstrating significant improvements in image-text retrieval performance. Visualization experiments can effectively assist us in observing the improvement of the model’s performance and the enhanced effects brought about by each added module,making the experimental results highly intuitive. Additionally,the proposed method maintains good retrieval performance in data robustness experiments. The average recall rate of the model consistently exceeds 70%,and it maintains excellent retrieval performance. This outcome indicates that the proposed model exhibits remarkable data robustness and is applicable to various scenarios,such as fuzzy retrieval.

Conclusion

The proposed method enhances the model’s ability to align Han portrait stone images with their descriptive texts. The model can learn effective and crucial features by simultaneously enhancing text and images. The proposed dynamic temperature mechanism adapts to the structural characteristics of the Han portrait stone dataset and enhances the model’s ability to learn from difficult samples. effectively improving the performance of image-text retrieval in terms of accuracy and comprehensiveness. However,when objects in images are similar but differ in positional information,the image-text retrieval results may suffer from confusion and false positives. Given the limited scale and diversity of the dataset for Han portrait stone image-text retrieval,the model’s cross-modal retrieval performance may be constrained. In the future,we will optimize the model’s capability to semantically analyze the different positions and actions of objects in images and expand the Han portrait stone dataset.

Han portrait stones  /  image-text retrieval  /  semantic enhancement  /  text reorganization  /  dynamic temperature
姜坤, 钱文华, 刘朋. 语义增强与文本重组协同的汉画像石图文检索. 中国图象图形学报, 2026 , 31 (8) : 3030 -3044 . DOI: 10.11834/jig.250393
Kun Jiang, Wenhua Qian, Peng Liu. Image-text retrieval for Han portrait stones through the synergy of semantic enhancement and text reorganization[J]. Journal of Image and Graphics, 2026 , 31 (8) : 3030 -3044 . DOI: 10.11834/jig.250393
图文检索,又称跨模态检索,是一种通过文本描述检索相关图像或通过图像检索相关文本描述的技术。其核心目标是消除图像与文本之间的语义鸿沟,实现跨模态数据的精准对齐(Smeulders等,2000)。图文检索的技术路径通常包含视觉特征提取、文本语义编码和跨模态对齐,通过残差网络(residual network,ResNet)(He等,2016)或视觉变换器(vision Transformer,ViT)(Dosovitskiy等,2021)提取图像的全局与局部特征,利用循环神经网络(rerrent neural network,RNN)(Lipton等,2015)或Transformer完成文本语义编码,再借助联合嵌入空间或注意力机制等方式实现跨模态特征匹配(尹奇跃 等,2021)。例如,在网购场景中,用户输入商品名称可以检索出相关图像;医学领域中,医院通过影像数据可以检索出相关诊断报告等(姜定和叶茫,2023)。
汉画像石是中国汉代特有的一种石刻艺术,是研究汉代社会、经济与文化的图像史书。如图1所示,图中两个人在敲鼓,两个人在奏乐,一个人在喝酒,描述的是一幅汉代的宴饮聚会图,阴线刻细腻流畅,浅浮雕立体感强,构图饱满且动感强烈,充满力量感与生命力。
汉画像石通过其多样丰富的内容展现了汉代的社会生产与风俗习惯,每一幅画作都蕴含了汉人的精巧构思和美好的愿景寄托。目前,针对汉画像石的研究主要集中在艺术价值(甘胜楠,2019)和历史意义(渠超和翟京襄,2009)等方面,缺乏对汉画像石图像和文本的数字化分类及检索(Chen等,2023)。因此,研究汉画像石图文检索技术,可以将分散的图文数据,如人物服饰、器物形制和场景布局等进行数字化分类,有利于用户通过关键词、图像特征(如车马出行、宴请奏乐)等快速定位相关材料,为汉代社会史、艺术史和科技史等研究提供技术支持(宋维涛 等,2025)。
图文检索需要提取出图像和文本的特征并计算出两个方向的相似度矩阵,然后按照相似度排序得出检索结果(Radford等,2021)。早期的图文检索方法基于手工设计的视觉特征,如颜色、纹理等与文本关键词进行匹配,该方法依赖人工标注和简单统计。Smeulders等人(2000)总结了早期的基于内容的图像检索(content-based image retrieval,CBIR)技术方法,该方法通过颜色直方图、Gabor纹理等全局特征或尺度不变特征变换(scale-invariant feature transform,SIFT)等局部特征描述图像,结合文本标签进行检索。该方法直观、准确度高,但是依赖人工检索,时间成本高、步骤繁杂,不满足实时性要求。
随着神经网络的发展,双塔模型在跨模态匹配中经历了从全局特征对齐到细粒度交互的演进,其技术路径可划分为3个阶段。在全局特征对齐阶段,Faghri等人(2018)提出的VSE(visual semantic embedding)模型具有代表性,该模型采用卷积神经网络(convolutional neural network,CNN)提取图像全局特征、RNN编码文本序列,通过三元组损失优化特征对齐。但受限于模态本质差异及模型架构局限,VSE无法实现图像与文本的细粒度对齐。之后,局部特征交互被提出,Lee等人(2018)提出的SCAN(stacked cross attention network)模型成为重要突破,其结合快速区域卷积神经网络(faster region-based convolutional neural network,Faster R-CNN)(Ren等,2015)提取图像区域特征,通过交叉注意力实现图像区域与文本单词的双向匹配,首次达成区域—单词级细粒度对齐,打破了全局特征的限制。不过,SCAN的双重注意力机制可能导致对齐不一致,且由于未引入大规模预训练,其对数据稀疏场景适应性较差。
此后,基于大规模图文数据的预训练模型成为主流。在对比学习与统一语义空间构建方面,Radford等人(2021)提出的CLIP(contrastive language-image pre-training)模型利用4亿图文对训练,通过对比学习构建统一语义空间,支持零样本检索,但其细粒度对齐能力有限,对中文等低资源语言支持不足。同期,Jia等人(2021)提出的ALIGN(automatic learning of inter-modality grounding)模型进一步验证了数据规模优先于数据质量,使用18亿噪声图文对训练,采用简化双塔结构提升训练效率,不过仍缺乏细粒度交互机制。Kim等人(2021)提出的ViLT(vision-and-language Transformer)模型则摒弃卷积与区域监督,将图像patch与文本token直接输入Transformer,首次实现图像与文本的“原生统一输入”,计算效率较SCAN提升10倍以上,但图像patch未结合语义信息,对复杂场景适应性不足。Wang等人(2022)提出的OmniVL(one foundation model for image-language and video-language tasks)模型采用了统一的Transformer编码器,先粗粒度对比全局特征再细粒度重排,通过“粗—细”两阶段对齐策略平衡效率与精度,显著提升检索任务的召回率与排序精度。Qin等人(2022)提出的Deep Evidential Learning模型,通过证据损失函数增强了模型对复杂数据的鲁棒性,首次将噪声建模融入跨模态对齐,解决了实际场景中数据标注质量参差不齐的问题。Zeng等人(2024)提出的X2-VLM(all-in-one pre-trained model for vision-language tasks)模型通过patch离散化提升细粒度检索能力,解决了ViLT仅依赖patch的局限。Li等人(2023)提出的BLIP(bootstrapping language-image pre-training)模型采用混合编解码器架构与CapFilt数据自举,通过ITC(image-text contrastive loss)和ITM(image-text matching loss)强化细粒度对齐,显著提升了细粒度对齐能力。Wang等人(2023)提出的双向图像文本编码器(bidirectional encoder representation from image and text,BEiT-3)模型以统一掩码建模替代对比学习,通过掩码图像patch和文本token实现语义对齐,提升了特征泛化性。Yang等人(2023)提出的Chinese-CLIP(Chinese contrastive language-image pre-training)模型则针对性解决中文场景语义鸿沟,构建2亿中文图文数据集,填补了跨语言泛化空白。Ge等人(2024)提出的视觉语义空间自强化网络3SHNet(visual semantic spatial self- highlighting network)模型进一步突破,融合物体语义与空间位置信息,突破了孤立区域对齐的局限,显著提升对场景化描述的匹配精度。
然而,上述方法应用于汉画像石图文检索时存在以下问题:1)小样本图文特征提取不足。当前的预训练大模型用数量较大的图文对进行训练,以确保模型的有效训练和优化。由于汉画像石样本的稀缺性和抽象性以及相似性,模型在训练过程难以提取到关键特征,导致图像与文本无法正确对齐,最终导致图文检索的效果较差。2)领域偏移问题。当前的多模态预训练大模型虽然在大规模公开数据集上表现出色,但描述未见领域的图像时,表现出明显的领域偏移问题。模型的先验知识会主导检索过程,导致模型无法将图像信息与文本正确对齐,进而检索到无关的图像或文本,出现幻觉现象。3)图像文本难以对齐。汉画像石图像中通常包含许多场景与元素,模型难以将图像的小场景与描述文本语义相对应,从而导致错检和漏检。
为此,本文提出语义增强与文本重组协同的汉画像石图文检索方法,主要贡献在于:1)建立了汉画像石图文检索数据集,包括神仙传说、车马出行以及市井生活等题材;2)提出语义增强模块,通过汉画像石掩膜图像计算对象权重,生成注意力与增强特征,经多阶段融合形成组合特征,提升图像特征表达力与鲁棒性;3)提出文本重组模块,依据掩膜图像提取语义对象,将其与文本嵌入随机模板生成增强文本,结构化描述文本,提高图文匹配度;4)引入动态温度机制,对高相似度样本设低温增强区分,低相似度样本设高温提升容错,对图像到文本、文本到图像双方向的相似度矩阵计算采用不同温度系数,平衡跨模态匹配难度,增强匹配准确性与稳定性。
本文提出的图文检索模型总体结构如图2所示。图文检索模型由图像编码器、文本编码器、语义增强模块和文本重组模块构成。图像端,将汉画像石图像与其掩膜图像统一尺寸后输入图像编码器,对输出的语义和图像特征归一化;依据文本和掩膜图像计算语义权重并归一化,结合语义权重与图像特征得出注意力权重,经残差连接生成注意力特征,同时将注意力权重作为增强特征,最终融合图像特征、注意力特征和增强特征得到组合特征,以丰富特征的表示。文本端,从掩膜图像提取不重复的有效语义对象,嵌入随机提示模板重组为增强文本,以规范文本表达并丰富文本语义信息,经编码器提取并归一化特征。然后计算组合特征与增强文本的特征相似度,推导温度参数生成相似度矩阵以适应本文数据集的结构,最后通过对比损失优化模型。
图像编码器使用ViT提取图像的特征,ViT通过自注意力机制直接建模图像分块之间的全局依赖关系,自注意力机制(Vaswani等,2017)允许模型动态聚焦于图像中的关键区域,避免了CNN因局部卷积操作导致的长距离语义断裂问题(Dosovitskiy等,2021)。此外,为了与文本编码器协同优化,本文采用ViT作为模型的图像编码器。
为了确保输入模型的图像大小一致,图像在送入图像编码器之前会被统一成384 × 384像素的大小,然后进行特征提取。
图3显示了图像编码器结构。图像进入编码器后首先被分块,将汉画像石图像转换为序列数据,使用16 × 16像素的patch,总共被分成了576个patch。而后对得到的patch进行向量化处理,每个patch通过线性投影层转换为向量,在降低复杂度的同时保留汉画像石的关键视觉信息,如线条、局部结构等。而后为每个patch添加位置编码,帮助模型理解汉画像石图像的空间结构,如图4车马出行图中人物、动物与马车之间的位置关系。将处理好的patch送入Transformer编码器层,编码器共有12层,每层包含多头自注意力、前馈神经网络、层归一化和残差连接模块,通过多个注意力头从线条方向、图案对称性等不同角度提取特征,增强特征的多样性。最后进行全局平均池化,对576个向量patch求平均值,得到一个1 × 512维的特征向量。
文本编码器使用RoBERTa(a robustly optimized bert pretraining approach)(Liu等,2019),可以捕捉文本中的关键语义信息,具有较高的特征提取效率,由于其使用的Transformer-Encoder层可以更好地捕捉词元之间的语义关联,可以从文本中提取更多有效的特征,且该编码器对中文数据集的适应性较好,因此,本文选用RoBERTa模型作为文本编码器。
如图5所示,文本输入进到文本编码器后,编码器首先会将文本按照词或字的粒度分割为词元序列。而后文本编码器会为每个词元分配一个向量表示,这个向量蕴含了词元的语义信息。与图像编码器类似,为每个词元在文本序列中的位置添加位置编码,这样可以让模型理解词元的先后顺序,最终生成768维的向量序列。该向量序列会被送至Transformer编码器层,该层包含了多头自注意力机制和前馈网络以及层归一化,通过自注意力捕捉每个词元之间的语义关联,理解文本的整体语义。前馈神经网络对自注意力输出的特征进行非线性变换,进一步增强特征的表达能力,区分出细粒度特征。最后通过跨模态投影层将768维向量线性变换为512维,与图像特征维度对齐,实现汉画像石跨模态检索。
为了让提取到的汉画像石图像特征中人物、动物等主体对象的特征更为突出,本文引入了语义增强模块,其结构如图6所示,根据掩膜图像所提供的语义信息对图像的特征进行增强,使那些关键主体的特征更加突出,并与动态温度机制结合,显著提升了模型的检索性能。
本文首先根据掩膜图像信息和文本信息计算语义权重,语义权重包括文本权重和图像权重两部分。文本权重的计算首先将该批次内的所有文本以空格隔开转换成字符串,然后统计总文本中该语义对象的出现次数,计算为
T_w=∑i=1Ut∈Tstr(t)I(Ut∈Tstr(t)i=n)Ut∈Tstr(t)
式中,T_w表示文本权重,n表示语义对象的名称,T表示所有文本,t表示单个文本。I为指示函数,括号内条件成立时取值为1,不成立则为0。U表示把批次内所有文本t拼接合并成一条完整长字符串。str表示拼接后长字符串的第i个词元/字符。
计算图像权重时,首先统计该批次内所有图像的颜色张量与当前输入语义对象的颜色张量相等的次数,然后得出该次数在该批次图像总像素数的占比,获得图像权重,计算为
I_w=∑b=1B∑h=1H∑w=1WI(∩p=13(Sb,p,h,w=O1,p,1,1))B⋅H⋅W
式中,I_w表示图像权重,B表示总批次大小,p表示通道信息,H、W表示图像高和宽,O表示原图像,S表示掩膜图像,b表示具体批次,h,w分别表示像素点的具体高和宽。
语义权重的计算基于每个对象的图像权重和文本权重,由于掩膜图像中黑色背景的占比较高,并且黑色背景属于无效的语义信息,所以在权重加和时,图像权重的占比仅设置为0.1,文本权重的占比设置为0.9,以确保计算出合理的语义权重,计算为
w[i]=0.9×T_w+0.1×I_w
式中,w表示语义权重。
通过计算得到各语义对象的权重,过滤掉了一些无关信息,让模型关注权重更高的对象。在此基础上,根据语义权重计算出注意力特征和增强特征,首先根据原始图像特征初始化注意力权重,进而根据掩膜图像张量和目标语义对象颜色张量计算出语义掩码,然后根据语义掩码在所有像素的占比计算出该掩码的重要性,计算为
an=∑h=1H∑w=1WI∩p=13(Sb,p,h,w=O1,p,1,1n)H⋅W
式中,
an
表示掩码的重要性。
根据计算所得掩码重要性和语义权重计算出注意力权重,对注意力权重残差连接再乘以掩膜图像的特征则得到了最终的注意力特征。计算式为
Fatt=softmax(∑n∈Nwn⋅(Fseg⊙an))⊙Fseg
式中,Fatt表示注意力特征,N表示语义对象集合,Fseg表示掩膜图像的特征,
⊙
是矩阵乘法,softmax是激活函数(Jang等,2017)。
基于注意力权重计算增强特征,去掉了残差连接的步骤,直接对特征进行增强,丰富特征的表示。将注意力权重直接作为增强特征,计算为
Fenh=∑n∈Nwn⋅(Fseg⊙an)
式中,Fenh表示增强特征。
该模块设计了多阶段特征融合,将原始图像特征、注意力特征和增强特征融合得到最终的组合特征,计算为
Fcomb=αFimg+βFatt+γFenh
式中,Fcomb表示组合特征,Fimg表示汉画像石图像特征,
α
、
β
、
γ
表示比例系数。
在得到组合特征后,计算其与增强文本特征的相似度。为了适应本文的特殊数据集结构,提出了动态温度调整和非对称相似度计算,模型会根据相似度高低自动调整对比学习的温度。当相似度较高时则使用较低的温度以增强区分度;相似度较低时则使用较高温度增加容错性,这样可以提升模型在不同难度的汉画像石样本上的学习效果,计算为
T=Tbase⋅[1+a⋅(1-1n∑i=1n(Fimg⋅FtextT))]
式中,T为温度,Tbase = 0.07,参考了CLIP的设置(Radford等,2021),
a
为激进系数,表示相似度对温度的影响强度,本文设置为0.3,Ftext表示文本特征。
由于汉画像石图像多数场景表现出相同的语义信息,因此本文数据集设计文本与图像是一对多的关系,图像与文本是一对一的关系。为了应对此结构,本模型为双向的检索设置了不同的温度系数以平衡两种方向的匹配难度,充分考虑图像和文本特征的分布差异,提高跨模态匹配的准确性和稳定性,计算为
Limg2text=Fimg⋅FtextT0.9TLtext2img=Ftext⋅FimgT1.1T
式中,L表示相似度矩阵,图像到文本和文本到图像检索的温度系数分别设置为0.9和1.1。
本文提出了文本重组模块,它可以充分利用掩膜图像中的有效语义信息并嵌入规范提示模板,使得汉画像石描述文本更加规范化并且包含更多语义信息,与图像特征增强协同,提升汉画像石图文检索的性能。其结构如图7所示。
该模块首先从输入图像对应的掩膜图像中提取出有效且不重复的语义对象,然后将这些对象作为Context,语义对象获取的过程计算为
V={n∈N|n∉{g, pad, cls, sep}∧¬∃n'∈N, n'=n}
式中,V表示有效语义对象,g表示背景,n’表示单个语义对象,pad(填充)、cls(分类标记)、sep(分隔符)等是特殊字符。
汉画像石描述文本进入该模块后,将为该文本选择一个随机提示模板,并将文本内容Text与包装后的语义对象信息Context嵌入到该模板中,计算式为
format(t, s, n)=t.replace{text→s, context→c}
式中,t是随机提示模板,s表示文本,“
→
”表示嵌入模板。
汉画像石的题材包罗万象、异彩纷呈,既生动描摹了古人的现实生活图景,又深刻承载了先民的神话信仰世界,更忠实记录了波澜壮阔的历史与声势浩荡的车马出行场景。
为了提高汉画像石图像的质量,本文从汉画像石博物馆和网络搜集整理素材。通过对原始质量不佳的汉画像石图像进行裁剪、清晰化和模糊部分消除等处理保留清晰可识别的部分,如图8所示。
此外,为了应对汉画像石图像数据集不足的问题,本文通过提取出汉画像石中的不同对象,如图9所示,将不同的对象组合拼凑成新的汉画像石图像,以此对汉画像石数据集进行扩充。
经过优化和扩充后的汉画像石图文检索数据集题材包含市井生活、神仙传说和车马出行,图像数量分别约是300、270和260幅,根据素材自建文本描述语句共479句。数据集部分示例如图10所示,其中一条文本可以对应多个图像场景的语义表达,如图中文本“四个女娲”可以对应两幅图像,单幅图像只能对应单条文本,如图中四人骑马出行图对应唯一文本“四个人骑着马出行”。
实验环境基于Pytorch的深度学习框架,显卡型号为Geforce NVIDIA RTX 4090,模型的图像编码器为ViT,文本编码器为RoBERTa,为加速汉画像石图文检索任务的收敛,各模型均采用了与其相对应的预训练模型进行微调,使用单卡训练,训练过程中权重衰减为0.01/10 epoch,batch_size设置为24,初始学习率设置为2 × 10-5,优化器为AdamW(Loshchilov和Hutter,2019),并使用交叉熵损失函数(Mao等,2023)对模型进行微调,通过对比损失(van den Oord等,2019)优化模型,实验数据集按照6∶2∶2的比例进行分配,其中,训练集包含280个文本和500幅图像,验证集包含95个文本和164幅图像,测试集包含104个文本和168幅图像。
为了验证模型的有效性,本文采用领域内常用的评估策略召回率Recall@k(Hawking等,1999),本文将k的值设定为1、5和10,通过计算图像到文本和文本到图像的Recall@1、Recall@5、Recall@10,进而反映检索结果的精确性、全面性和鲁棒性。Recall@k通过计算前k个结果中真正相关的数量占比来评估模型的检索性能,计算式为
Recall@k=TP@kTP@k+FN@k
式中,FN@k表示未被检索到的相关实例数量(即漏检),TP@k表示前k个结果检索到的数量。
本文对特征融合参数进行讨论,如表1所示,表中对不同特征比例组合下的图文检索指标进行了对比。为了保留原始特征占有一定的比重,本实验设置原始特征最低为0.4、最高为0.8,根据不同的特征比例计算比较评价指标,最终得出原始特征、注意力特征和增强特征按照0.4、0.5和0.1的比例是最佳的特征融合方案,在图像到文本检索和文本到图像检索的实验中都获得最高的指标。通过实验可知,以注意力特征和原始特征为主导、增强特征为辅助的搭配检索指标最高,原始特征保留较多则会导致模型性能下降,注意力特征的加入让模型可以更好地识别汉画像石图像中的关键部分。
为了验证本文模型的有效性,选取BLIP、ViLT、VSE、X2-VLM、BEiT3、Chinese-CLIP、DSMD和3SHNet等模型进行比较。本文对不同模型的图文检索性能进行了评估,对比结果如表2所示,从表中可以看出,本文模型的图像检索评价指标Mean_Recall、Recall@1、Recall@5和Recall@10在文本到图像的检索中分别达到79.48%、55.76%、87.5%和95.19%,在图像到文本的检索中分别达到78.37%、52.38%、85.71%和97.02%,在精确性和全面性上,文本到图像的检索指标Mean_Recall、Recall@1、Recall@5和Recall@10相较于排名第2的BLIP模型分别提高15.20%、21.84%、14.29%和9.48%。图像到文本的检索指标Mean_Recall、Recall@1、Recall@5和Recall@10相较于排名第2的X2-VLM模型分别提高17.47%、23.22%、18.45%和10.72%。可以看出,本文所提模型在文本到图像检索和图像到文本检索的精确性和全面性都得到了提升,解决了领域偏移问题,图文特征能更好地进行对齐。
为了验证本文提出的语义增强模块、动态温度模块以及文本重组模块的有效性,本文设置了相应的消融模型,实验结果见表3。
模型首先通过添加动态温度模块提高了模型跨模态检索的稳定性和准确性。在此基础上,引入多阶段特征融合模块,使模型关注图像中的重要区域,提取更多关键的特征。最后,引入文本重组模块与多阶段特征融合模块协同以增强模型对齐汉画像石图像与文本特征的能力。实验结果显示,在图文检索的精确性和全面性方面,本文模型相较于其他模型都有一定程度的提升,相较于不添加任何模块的模型,文本到图像检索中,本文模型在图文检索指标Recall@1、Recall@5和Recall@10分别提升41.34%、47.12%和42.31%;图像到文本检索中,本文模型在图文检索指标Recall@1、Recall@5和Recall@10分别提升41.08%、55.36%和50.00%。结合消融实验结果看,本文添加的每个模块对模型图文检索的效果具有不同程度的增益作用,共同促进了图文检索性能。
此外,为了可视化模块添加对于模型效果的增益,针对此进行了正负样本分离度可视化实验,如图11所示。从图中可以看出,初始模型的正负样本存在较多重叠,峰值距离较近,区分度较低,仅添加多阶段特征融合模块和仅添加文本重组模块的模型相较于初始模型正负样本重叠区域有所减小,并且峰值距离拉远。集成了两个模块的模型正负样本的重叠区域最小,并且峰值距离最远,模型的区分度效果最好,达到0.93。
为了更直观地体现出本文模型在汉画像石图文检索任务中性能的提升,通过可视化实验反映模块添加对本文模型的增益效果,如图12所示。图12(a)显示了文本到图像的检索结果,图中实线边框表示该文本对应的正确检索图像,从初始前10个结果检索不到的情况,到进行微调、单图像增强和单文本增强,模型在全面性和精确性方面都得到了提升,本文提出的模型在检索的结果中前6个都是正确的,相较于其他模型,本文提出的模型可以捕捉到不同图像之间的同一语义表达,使得检索结果更加精确和全面。图12(b)显示了图像到文本的检索结果,加粗字体的文本表示该图像对应的正确描述文本,从初始模型前10个结果检索不到正确文本的情况,到进行微调、单图像增强和单文本增强,模型的性能得到了不错的提升,可以在前5个检索结果中检索到正确文本。本文提出的模型将图像增强与文本增强结合,提取到更丰富的图像特征和文本特征,模型能更好地将图文特征进行对齐,在检索效果上得到了不错的提升,可以达到检索到的正确结果中排名第一。
为了验证本文模型的鲁棒性,对于图像,采取了随机裁剪、添加高斯噪声、模糊化与亮度变化等随机操作,对于文本,采取了同义词替换、顺序变换等操作,在此情况下测试模型的检索性能,得到的指标对比如表4所示。虽然指标相较于原始数据有所下降,但是平均召回率始终可以保持不低于70%,模型依旧保持较好的检索性能,具有较好的鲁棒性。
本文进行了鲁棒性可视化实验,以验证鲁棒性,如图13所示,图像到文本的检索性能相较于初始数据有所衰减,但是模型依旧可以在前10个检索结果内检索出正确的结果。文本到图像的检索,模型依旧能够保持良好的精确性与全面性,在有所干扰的情况下依旧可以检索出多个准确的结果。通过该实验结果分析可以得知本文模型具有良好的鲁棒性。
针对汉画像石样本较少、内容丰富多彩的特点,本文提出了一种语义增强与文本重组协同的汉画像石图文检索模型。引入了语义增强模块,通过掩膜图像提供的语义信息计算得出注意力特征和增强特征,通过多阶段特征融合得到组合特征,在保留原始特征的基础上又突出关键特征。此外,基于本文数据集的特殊性设置了相应的温度模块,通过图文相似度计算温度,提高了跨模态检索的全面性与准确性。通过引入文本重组模块,利用从掩膜图像中提取出的有效语义信息和随机规范提示模板对文本进行重组,这样既丰富了文本的语义表达,又增强了文本的规范性。实验表明,本文模型较原始模型在精确性和全面性以及鲁棒性上有显著提升。然而,当图像中的对象比较相似但方位信息不一致时,图文检索的效果可能会出现混淆与误检。此外,由于汉画像石图文检索的数据集的规模与多样性有限,模型跨模态检索的性能可能会受到约束。未来,将扩充汉画像石数据集,并且优化模型对图像中对象的不同位置与动作进行语义分析的能力。
  • 国家自然科学基金项目(62162065);云南大学“双一流”建设联合专项(202401BF070001-023);云南省“视觉与文化计算创新团队”项目(202505AS350009);云南省科技厅应用基础研究计划项目(202201AT070167)
参考文献 引证文献
排序方式:
Chen W,Liu Y,Wang W P,Bakker E M,Georgiou T,Fieguth P,et al. 2023. Deep learning for instance retrieval: a survey. IEEE Transactions on Pattern Analysis and Machine Intelligence,45(6): 7270-7292 [DOI: 10.1109/TPAMI.2022.3218591]
Dosovitskiy A,Beyer L,Kolesnikov A,Weissenborn D,Zhai X H,Unterthiner T,et al. 2021. An image is worth 16 × 16 words: transformers for image recognition at scale//Proceedings of the 9th International Conference on Learning Representations. [s.l.]: #ICLR
Faghri F,Fleet D J,Kiros J R and Fidler S. 2018. VSE++: improving visual-semantic embeddings with hard negatives//Proceedings of 2018 British Machine Vision Conference. Newcastle,UK: BMVA Press: #12
Gan S N. 2019. Study on the Dance Art in the Stone Relief of the Han Dynasty in Nanyang from the Perspective of Imagolog. Zhengzhou: Zhengzhou University
甘胜楠. 2019. 图像学视域下南阳汉画像石中的舞蹈艺术研究. 郑州: 郑州大学
Ge X R,Xu S P,Chen F H,Wang J,Wang G X,An S,et al. 2024. 3SHNet: boosting image-sentence retrieval via visual semantic-spatial self-highlighting. Information Processing and Management,61(4): #103716 [DOI: 10.1016/j.ipm.2024.103716]
Hawking D,Voorhees E M,Craswell N and Bailey P. 1999. Overview of the TREC-8 web track//Proceedings of 1999 Eighth Text Retrieval Conference. Gaithersburg,USA: NIST
He K M,Zhang X Y,Ren S Q and Sun J. 2016. Deep residual learning for image recognition//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas,USA: IEEE: 770-778 [DOI: 10.1109/CVPR.2016.90]
Jang E,Gu S X and Poole B. 2017. Categorical reparameterization with gumbel-softmax//Proceedings of the 5th International Conference on Learning Representations. Toulon,France: ICLR
Jia C,Yang Y F,Xia Y,Chen Y T,Parekh Z,Pham H,et al. 2021. Scaling up visual and vision-language representation learning with noisy text supervision//Proceedings of the 38th International Conference on Machine Learning. [s.l.]: JMLR.org: 4904-4916
Jiang D and Ye M. 2023. Transformer network for cross-modal text-to-image person re-identification. Journal of Image and Graphics,28(5): 1384-1395
姜定,叶茫. 2023. 面向跨模态文本到图像行人重识别的Transformer网络. 中国图象图形学报,28(5): 1384-1395 [DOI: 10.11834/jig.220620]
Kim W,Son B and Kim I. 2021. ViLT: vision-and-language transformer without convolution or region supervision//Proceedings of the 38th International Conference on Machine Learning. [s.l.]: JMLR.org: 5583-5594
Lee K H,Chen X,Hua G,Hu H D and He X D. 2018. Stacked cross attention for image-text matching//Proceedings of the 15th European Conference on Computer Vision. Munich,Germany: Springer: 212-228 [DOI: 10.1007/978-3-030-01225-0_13]
Li J N,Li D X,Savarese S and Hoi S. 2023. BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models//Proceedings of the 40th International Conference on Machine Learning. Honolulu,USA: JMLR.org: #814
Liang Z Y,Liang M Y,Huang W,Li Y W,Liu W,Shao Y X,et al. 2025. Dynamic self-adaptive multiscale distillation from pre-trained multimodal large model for efficient cross-modal retrieval//Proceedings of the 33rd ACM International Conference on Multimedia. Dublin,Ireland: ACM: 2851-2859 [DOI: 10.1145/3746027.3754844]
Lipton Z C,Berkowitz J and Elkan C. 2015. A critical review of recurrent neural networks for sequence learning [EB/OL]. [2025-10-28]. https://arxiv.org/pdf/1506.00019.pdf
Liu Y H,Ott M,Goyal N,Du J F,Joshi M,Chen D Q,et al. 2019. RoBERTa: a robustly optimized BERT pretraining approach [EB/OL]. [2025-10-28]. https://arxiv.org/pdf/1907.11692.pdf
Loshchilov I and Hutter F. 2019. Decoupled weight decay regularization//Proceedings of the 7th International Conference on Learning Representations. New Orleans,USA: ICLR
Mao A Q,Mohri M and Zhong Y T. 2023. Cross-entropy loss functions: theoretical analysis and applications//Proceedings of the 40th International Conference on Machine Learning. Honolulu,USA: JMLR.org: 23803-23828
Qin Y,Peng D Z,Peng X,Wang X and Hu P. 2022. Deep evidential learning with noisy correspondence for cross-modal retrieval//Proceedings of the 30th ACM International Conference on Multimedia. Lisboa,Portugal: ACM: 4948-4956 [DOI: 10.1145/3503161.3547922]
Qu C and Zhai J X. 2009. Analysis of Han dynasty stone reliefs in Xuchang. Legal System and Society,(7): #311
渠超,翟京襄. 2009. 浅析许昌汉画像石. 法制与社会,7: #311 [DOI: 10.19387/j.cnki.1009-0592.2009.07.201]
Radford A,Kim J W,Hallacy C,Ramesh A,Goh G,Agarwal S,et al. 2021. Learning transferable visual models from natural language supervision//Proceedings of the 38th International Conference on Machine Learning. [s.l.]: JMLR.org: 8748-8763
Ren S Q,He K M,Girshick R and Sun J. 2015. Faster R-CNN: towards real-time object detection with region proposal networks//Proceedings of the 29th International Conference on Neural Information Processing Systems. Montreal,Canada: MIT Press: 91-99
Smeulders A W M,Worring M,Santini S,Gupta A and Jain R. 2000. Content-based image retrieval at the end of the early years. IEEE Transactions on Pattern Analysis and Machine Intelligence,22(12): 1349-1380 [DOI: 10.1109/34.895972]
Song W T,Liao L Y,Zhang H T,Li L,Yu T X,Zhao Y S,et al. 2025. Applications and prospects of artificial intelligence in the cultural heritage. Journal of Image and Graphics,30(12): 3707-3739
宋维涛,廖聆宇,张浩天,李琳,俞天秀,赵永生,等. 2025. 人工智能在文物行业的应用与展望. 中国图象图形学报,30(12): 3707-3739 [DOI: 10.11834/jig.240765]
van den Oord A,Li Y Z and Vinyals O. 2019. Representation learning with contrastive predictive coding [EB/OL]. [2025-10-28]. https://arxiv.org/pdf/1807.03748.pdf
Vaswani A,Shazeer N,Parmar N,Uszkoreit J,Jones L,Gomez A N,et al. 2017. Attention is all you need//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach,USA: Curran Associates Inc.: 6000-6010
Wang J K,Chen D D,Wu Z X,Luo C,Zhou L W,Zhao Y C,et al. 2022. OmniVL: one foundation model for image-language and video-language tasks//Proceedings of the 36th International Conference on Neural Information Processing Systems. New Orleans,USA: Curran Associates Inc.: #412
Wang W H,Bao H B,Dong L,Bjorck J,Peng Z L,Liu Q,et al. 2023. Image as a foreign language: BEIT pretraining for vision and vision-language tasks//Proceedings of 2023 Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver,Canada: IEEE: 19175-19186 [DOI: 10.1109/CVPR52729.2023.01838]
Yang A,Pan J S,Lin J Y,Men R,Zhang Y C,Zhou J R,et al. 2023. Chinese CLIP: contrastive vision-language pretraining in Chinese [EB/OL]. [2025-10-28]. https://arxiv.org/pdf/2211.01335.pdf
Yin Q Y,Huang Y,Zhang J G,Wu S and Wang L. 2021. Survey on deep learning based cross-modal retrieval. Journal of Image and Graphics,26(6): 1368-1388
尹奇跃,黄岩,张俊格,吴书,王亮. 2021. 基于深度学习的跨模态检索综述. 中国图象图形学报,26(6): 1368-1388 [DOI: 10.11834/jig.200862]
Zeng Y,Zhang X S,Li H,Wang J W,Zhang J P and Zhou W C S. 2024. X2-VLM: all-in-one pre-trained model for vision-language tasks. IEEE Transactions on Pattern Analysis and Machine Intelligence,46(5): 3156-3168 [DOI: 10.1109/TPAMI.2023.3339661]
2026年第31卷第8期
PDF下载
11
5
引用本文
BibTeX
文章信息
doi: 10.11834/jig.250393
  • 接收时间:2025-10-28
  • 首发时间:2026-10-30
  • 出版时间:2026-08-16
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2025-10-28
  • 修回日期:2026-01-15
基金
National Natural Science Foundation of China(62162065);Joint Special Project Research Foundation of Yunnan Province(202401BF070001-023);Yunnan Province Visual and Cultural Innovation Team(202505AS350009);Yunnan Fundamental Research Projects(202201AT070167)
国家自然科学基金项目(62162065);云南大学“双一流”建设联合专项(202401BF070001-023);云南省“视觉与文化计算创新团队”项目(202505AS350009);云南省科技厅应用基础研究计划项目(202201AT070167)
作者信息
    云南大学信息学院,昆明 650504

通讯作者:

钱文华,通信作者,男,教授,博士生导师,主要研究方向为计算机视觉和文化计算。E-mail:
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/10.11834/jig.250393
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏