Scholarlink

<p>Image hiding is a field that focuses on data transmission and storage. With the development of technology, the information hiding capacity is increasing and the container image can hide one or several secret images. However, for container images, higher hiding capacity means higher risk. Currently, image hiding algorithms based on Invertible Neural Networks (INN) will discard some of the useful information during the hiding process, which will lead to the quality degradation of the secret image pair. Especially for the multi-image hiding process, the quality of the image decreases drastically. To solve this problem, we introduce a new image hiding framework called Mapping model based on Invertible Network (MapIN). Specifically, a mapping module is designed to simulate the discarded loss information during the hiding process, therefore the loss of information during the revealing process can be reduced, and the secret image will be extracted from the stego image losslessly. To further improve the security and invisibility of steganography methods, we introduce a deep perceptual loss in the loss function. The PSNR values of the secret/recovery image pair are 48.87&#xa0;dB, 45.98&#xa0;dB, and 42.48&#xa0;dB on the DIV2K, ImageNet and COCO dataset, respectively, and the accuracies are close to 50% (random guessing state) under the detection of the steganalyser SRNet, Zhu-Net and SiaStegNet. The experimental results show that our method outperforms other State-Of-The-Art (SOTA) image hiding methods on different image datasets, and it can also work well in the field of multi-images hiding.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mapping model based on invertible networks for image hiding

  • 期刊论文

  • 发表日期：

    2025年1月28日

查看全文
  • Yuwei Wang,
  • Yan Zhao,
  • Han Liang,
  • Liang Xue

摘要

Image hiding is a field that focuses on data transmission and storage. With the development of technology, the information hiding capacity is increasing and the container image can hide one or several secret images. However, for container images, higher hiding capacity means higher risk. Currently, image hiding algorithms based on Invertible Neural Networks (INN) will discard some of the useful information during the hiding process, which will lead to the quality degradation of the secret image pair. Especially for the multi-image hiding process, the quality of the image decreases drastically. To solve this problem, we introduce a new image hiding framework called Mapping model based on Invertible Network (MapIN). Specifically, a mapping module is designed to simulate the discarded loss information during the hiding process, therefore the loss of information during the revealing process can be reduced, and the secret image will be extracted from the stego image losslessly. To further improve the security and invisibility of steganography methods, we introduce a deep perceptual loss in the loss function. The PSNR values of the secret/recovery image pair are 48.87 dB, 45.98 dB, and 42.48 dB on the DIV2K, ImageNet and COCO dataset, respectively, and the accuracies are close to 50% (random guessing state) under the detection of the steganalyser SRNet, Zhu-Net and SiaStegNet. The experimental results show that our method outperforms other State-Of-The-Art (SOTA) image hiding methods on different image datasets, and it can also work well in the field of multi-images hiding.