基于改进ConvNeXtBlock的新型双域融合图像隐写

打开文本图片集

关键词：图像隐写；深度可分离卷积；空间域；频域；安全性；大容量

中图分类号：TN9.73；TP309.2 文献标志码：A 文章编号：67-684（2026）0-000-09

DOI： 0. 3705/j. issn.67-684.202423

Abstract：To address the poor imperceptibility， insufficient security，and low steganographic capacity in image steganography，a novel dual-domain fusion image steganography scheme based on improved ConvNeXt Block was proposed.Firstly，the improved depth separable convolution module could learn more detailed image feature information. Secondly，a novel spatial and frequency domain information fusion way was designed to improve image imperceptibility and security. Finally，multiple loss functions were employed to constrain the network in a cascaded manner. Experimental results showed that the proposed scheme achieved an average increase of 3-4 dB in peak signal-to-noise ratio compared to other steganography schemes，and the average structural similarity and learned perceptual image patch similarity were 0. 99 and O. OO1， respectively. The resistance to steganalysis was closer to 50% ，indicating higher security. Even with high-capacity hiding，good results were still maintained.

Key words： image steganography； depth separable convolution; spatial domain； frequency domain； se-curity; high-capacity

0 引言

图像隐写是一种用于保护秘密图像的技术，其实现过程是将一幅秘密图像嵌入另一幅载体图像中，使生成的载密图像与原始载体图像之间无法区分，同时要求接收方能够从载密图像中成功地提取出隐藏的秘密图像。（剩余12205字）