Asymmetric cross–modal hashing with high–level semantic similarity. (October 2022)

• Record Type: Journal Article
• Authors: Yang, Fan; Liu, Yufeng; Ding, Xiaojian; Ma, Fumin; Cao, Jie
• Is Part Of: Pattern recognition. Volume 130(2022)
• Publication Date: 2022-10
• Subjects: Cross-modal retrieval -- Hashing -- Similarity search -- Supervised -- Optimization; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108823
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: To give reliable supervisory signals for the hash learning procedure, we develop a simple but effective supervised cross-modal hash learning framework to learn more discriminative hash codes and preserve the semantic similarity. An effective iterative alternative optimization scheme is developed to solve the NP-hard optimization problem, the time consuming and space complexity of our approach is O(n). We propose a novel cross-modal hashing approach, which can effectively embed high level semantic similarity into the learning process of hash codes. A two-stage hash coding learning strategy is designed to optimize learning hash codes and hash functions respectively; our strategy effectively reduces the information loss in the process of instances mapping hash codes. To some extent, it improves the hash code learning rate of new samples, so our strategy can be extended to large-scale multi-mode information retrieval. We propose a novel semantic-enhanced scheme to make full leverage the label information and gain more powerful hash functions, our approach can generate more discriminative hash codes for new instances via powerful hash functions. we conduct comprehensive experiments on several datasets, experimental results demonstrate the superior effectiveness of our approach outperforms several state-of-the-art hashing models on either the retrieval accuracy or the hash learning efficiency. Abstract: Cross-modal hashing aims at using modality content to retrieve semantically relevant objects of different modalities, so cross-modal retrieval has attracted much attention. To effectively exploit the discriminative label information and retain more semantic information in the process of hash learning, we propose a novel cross-modal hashing method, named high-level semantic similarity analysis hashing (HSSAH) for cross-modal retrieval. To reduce time complexity and enhance discriminant ability in hash codes, HSSAH constructs an asymmetric high-level semantic similarity learning framework to replace the binary semantic similarity matrix. Moreover, the developed HSSAH is a two-stage approach, and a semantic-enhanced scheme is proposed in the second stage, which fully leverages the label information to gain more powerful hash functions. We conducted comprehensive experiments on three benchmark datasets to evaluate the performance of HSSAH. Experimental results show that HSSAH can achieve significantly better retrieval precision and outperforms several state-of-the-art approaches.