View in EDS

Multi-image steganography via function extension of implicit neural representation.

Saved in:
Bibliographic Details
Title: Multi-image steganography via function extension of implicit neural representation.
Authors: Lu, Yuwei1,2 (AUTHOR) 1349373547@qq.com, Liu, Jia1,2 (AUTHOR) liujia1022@gmail.com, Wang, Qiya1,2 (AUTHOR) 1577620009@qq.com, Liu, Yujie1,2 (AUTHOR) 17758142958@163.com, Luo, Peng1,2 (AUTHOR) lp_nwpu@mail.nwpu.edu.cn
Source: Multimedia Systems. Jun2026, Vol. 32 Issue 4, p1-18. 18p.
Subjects: Implicit functions, Artificial neural networks, Image transmission, Cryptography
Abstract: As deep learning technology undergoes rapid evolution, to address the demand for larger-capacity image steganography, researchers have proposed many steganography methods based on deep neural network (DNN), leveraging DNN's powerful representation ability. However, DNN-based multi-image steganography schemes suffer from issues such as inconvenient decoder transmission, narrow applicability, and high training resource demands. In recent years, implicit neural representation (INR) has been applied to image steganography, enhancing the concealment of steganographic image transmission. This paper proposes an implicit neural representation-based multi-image steganography method leveraging a functional continuity extension strategy, aiming to address issues in deep learning-based multi-image steganography such as high training costs, heavy decoder transmission overhead, and resolution constraints, while further improving steganographic capacity and efficiency. Implicit neural representations encode data as continuous functions and can effectively handle data across multiple modalities and resolutions. In this work, we represent images as continuous functions using implicit neural representations and employ a strategy of functional continuity extension and nesting, controlled by keys, to hierarchically embed multiple secret images into the stego-function corresponding to a cover image. Experimental results show that, on the COCO dataset, when three secret images are embedded into a single cover, the average PSNR of the recovered secret images reaches 27.11 dB, demonstrating a relatively high embedding capacity and training efficiency under the premise of acceptable visual quality and security. [ABSTRACT FROM AUTHOR]
Copyright of Multimedia Systems is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:As deep learning technology undergoes rapid evolution, to address the demand for larger-capacity image steganography, researchers have proposed many steganography methods based on deep neural network (DNN), leveraging DNN's powerful representation ability. However, DNN-based multi-image steganography schemes suffer from issues such as inconvenient decoder transmission, narrow applicability, and high training resource demands. In recent years, implicit neural representation (INR) has been applied to image steganography, enhancing the concealment of steganographic image transmission. This paper proposes an implicit neural representation-based multi-image steganography method leveraging a functional continuity extension strategy, aiming to address issues in deep learning-based multi-image steganography such as high training costs, heavy decoder transmission overhead, and resolution constraints, while further improving steganographic capacity and efficiency. Implicit neural representations encode data as continuous functions and can effectively handle data across multiple modalities and resolutions. In this work, we represent images as continuous functions using implicit neural representations and employ a strategy of functional continuity extension and nesting, controlled by keys, to hierarchically embed multiple secret images into the stego-function corresponding to a cover image. Experimental results show that, on the COCO dataset, when three secret images are embedded into a single cover, the average PSNR of the recovered secret images reaches 27.11 dB, demonstrating a relatively high embedding capacity and training efficiency under the premise of acceptable visual quality and security. [ABSTRACT FROM AUTHOR]
ISSN:09424962
DOI:10.1007/s00530-026-02242-9