Multi-head similarity feature representation and filtration for image-text matching

Date

2023

Authors

Jiang, M.
Zhang, S.
Cheng, D.
Zhang, L.
Zhang, G.

Editors

Advisors

Journal Title

Journal ISSN

Volume Title

Type:

Book chapter

Citation

Event/exhibition information: 19th International Conference on Advanced Data Mining and Applications, ADMA 2023, Shenyang, 21/08/2023-23/08/2023 Source details - Title: International Conference on Advanced Data Mining and Applications ADMA 2023: Advanced Data Mining and Applications, 2023, vol.14177 LNAI, pp.629-643

Statement of Responsibility

Conference Name

Abstract

The field of multimedia analysis has been increasingly focused on image-text retrieval, which aims to retrieve semantically relevant images or text through queries of the opposite modality. The key challenge is to learn the correspondence between images and text. Existing methods have focused on processing inter-modality information interaction but have not given sufficient attention to learning the correspondence between the two modalities during this process. However, these methods have a low accurate image-text matching due to they are not deal with the noise during the process of the visual and textual representations. To avoid the noise in the training process, we propose a novel Multi-head Similarity Feature Representation and Filtration (MSFRF) approach for image-text matching. The proposed MSFRF method captures the detailed associations of feature representations from different modalities and reduces the interference of noisy information in the extracted features for improving the performance of matching. Extensive experiments on two benchmark datasets show that the proposed MSFRF method outperforms the state-of-the-art image-text matching methods.

School/Discipline

Dissertation Note

Provenance

Description

Access Status

Rights

Copyright 2023 The Author(s), under exclusive license to Springer Nature Switzerland. Access Condition Notes: Accepted manuscript available after 1 January 2025

License

Grant ID

Call number

Persistent link to this record