Shu-Juan Peng , Xueting Jiang , Xin Liu , Ji-Xiang Du , Jianjia Cao
{"title":"LPGOH: Label-prototype guided online hashing for efficient cross-modal retrieval","authors":"Shu-Juan Peng , Xueting Jiang , Xin Liu , Ji-Xiang Du , Jianjia Cao","doi":"10.1016/j.knosys.2025.114159","DOIUrl":null,"url":null,"abstract":"<div><div>Cross-modal hashing has recently received widespread attention due to its fast query speed, and existing batch-based methods are generally inefficient in an online scenario, i.e., multi-modal data points appear in a streaming manner. Although some online cross-modal hashing methods have been explored, they often neglect the semantic interdependency among the label categories and potentially suffer from the limited semantic preservations between newly coming data and existing data. To alleviate this concern, this paper proposes an efficient label-prototype guided online hashing (LPGOH) for cross-modal retrieval, which can incrementally learn the discriminative hash codes of streaming data while adaptively optimizing the hash function in a streaming manner. To be specific, the proposed framework first innovates a group of label-prototype codes to exploit the semantic interdependency between the label categories, and then combine the semantic similarity regularization to jointly learn the semantic-preserving hash codes. Meanwhile, <span><math><mrow><mi>ε</mi></mrow></math></span>-dragging operation is seamlessly utilized to provide provable large semantic margins, which can further promote the discrimination power of the learnt hash code and speed up the learning process. Besides, an online discrete optimization algorithm is efficiently designed to parse the semantic interdependency between the label categories, learn the compact hash codes for the current arriving data, and optimize the hash functions adaptively. Accordingly, the hash codes of streaming data are discriminatively learned to benefit various online cross-modal retrieval tasks. Extensive experiments evaluated on benchmark datasets verify the advantages of the proposed LPGOH framework, by achieving the competitive and mostly improved retrieval performance over the state-of-the-arts.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"327 ","pages":"Article 114159"},"PeriodicalIF":7.6000,"publicationDate":"2025-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Knowledge-Based Systems","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0950705125012006","RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0
Abstract
Cross-modal hashing has recently received widespread attention due to its fast query speed, and existing batch-based methods are generally inefficient in an online scenario, i.e., multi-modal data points appear in a streaming manner. Although some online cross-modal hashing methods have been explored, they often neglect the semantic interdependency among the label categories and potentially suffer from the limited semantic preservations between newly coming data and existing data. To alleviate this concern, this paper proposes an efficient label-prototype guided online hashing (LPGOH) for cross-modal retrieval, which can incrementally learn the discriminative hash codes of streaming data while adaptively optimizing the hash function in a streaming manner. To be specific, the proposed framework first innovates a group of label-prototype codes to exploit the semantic interdependency between the label categories, and then combine the semantic similarity regularization to jointly learn the semantic-preserving hash codes. Meanwhile, -dragging operation is seamlessly utilized to provide provable large semantic margins, which can further promote the discrimination power of the learnt hash code and speed up the learning process. Besides, an online discrete optimization algorithm is efficiently designed to parse the semantic interdependency between the label categories, learn the compact hash codes for the current arriving data, and optimize the hash functions adaptively. Accordingly, the hash codes of streaming data are discriminatively learned to benefit various online cross-modal retrieval tasks. Extensive experiments evaluated on benchmark datasets verify the advantages of the proposed LPGOH framework, by achieving the competitive and mostly improved retrieval performance over the state-of-the-arts.
期刊介绍:
Knowledge-Based Systems, an international and interdisciplinary journal in artificial intelligence, publishes original, innovative, and creative research results in the field. It focuses on knowledge-based and other artificial intelligence techniques-based systems. The journal aims to support human prediction and decision-making through data science and computation techniques, provide a balanced coverage of theory and practical study, and encourage the development and implementation of knowledge-based intelligence models, methods, systems, and software tools. Applications in business, government, education, engineering, and healthcare are emphasized.