Hiding images in audio based on invertible neural networks

Invertible Neural Networks(INNs) are well suited for the field of information hiding due to the fact that their inherent reversible structure. Images are able to efficiently convey information in a vivid and hierarchical manner, while audio is a widely used and distributed media file with a large embedding capacity. Therefore, hiding images in audio is of high research and application value. In the task of hiding images in audio, how to represent audio and image data and how to improve the quality of reconstructed images while reducing audio distortion are two important issues. To address these two problems, this paper proposes an algorithm based on INNs to hide images in audio. Inspired by the data processing methods in JPEG image compression, an image feature extraction and representation module is proposed for data feature representation. This module performs block-wise discrete cosine transform, Zigzag scanning, and high-low frequency separation operation on color images, extracting the frequency domain features of the image and obtaining its one-dimensional representation. In addition, in order to reduce audio distortion and improve the quality of reconstructed images, this paper uses the wavelet transform to separate the high and low frequency components of audio and introduces INNs to embed the secret image into the high-frequency region of the cover audio. Experimental results show that the proposed algorithm can generate higher quality steganographic audio and reconstruct more restored color images while achieving a high embedding rate, and that the proposed algorithm exhibits good security. © 2024 Science Press. All rights reserved.