Scholarlink

<p>With the widespread use of the Internet, the secure transmission and storage of image data have become increasingly critical due to the sensitive personal and commercial information often contained in images. Reversible Data Hiding in Encrypted Images (RDHEI) has emerged as a promising solution to simultaneously ensure image privacy and enable additional functionalities such as authentication and retrieval. However, existing RDHEI techniques face limitations in embedding capacity and adaptability to various image textures. This paper proposes a novel framework, ConcealNet, which extends Baluja’s deep learning-based image embedding method by integrating Denoising Diffusion Implicit Models (DDIM). In this approach, DDIM is first used to encrypt the cover image by transforming it into structured noise, enhancing the security of the encryption process. A deep neural network then embeds a secret image of the same size into the encrypted cover image, overcoming the hiding capacity limitations of traditional RDHEI methods. Experimental results demonstrate that ConcealNet achieves high visual quality in both decrypted and extracted images, with PSNR values of 52.82 dB and 78.38 dB, respectively. This framework significantly advances image privacy protection by offering a high-capacity, secure, and efficient embedding solution, showing strong potential for applications in secure image sharing and cloud storage environments.</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

ConcealNet: A Deep Learning Framework for Image Encryption and High-Capacity Data Embedding Using Denoising Diffusion Implicit Models

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Yi-Hui Chen,
  • Meng-Yu Tsai

摘要

With the widespread use of the Internet, the secure transmission and storage of image data have become increasingly critical due to the sensitive personal and commercial information often contained in images. Reversible Data Hiding in Encrypted Images (RDHEI) has emerged as a promising solution to simultaneously ensure image privacy and enable additional functionalities such as authentication and retrieval. However, existing RDHEI techniques face limitations in embedding capacity and adaptability to various image textures. This paper proposes a novel framework, ConcealNet, which extends Baluja’s deep learning-based image embedding method by integrating Denoising Diffusion Implicit Models (DDIM). In this approach, DDIM is first used to encrypt the cover image by transforming it into structured noise, enhancing the security of the encryption process. A deep neural network then embeds a secret image of the same size into the encrypted cover image, overcoming the hiding capacity limitations of traditional RDHEI methods. Experimental results demonstrate that ConcealNet achieves high visual quality in both decrypted and extracted images, with PSNR values of 52.82 dB and 78.38 dB, respectively. This framework significantly advances image privacy protection by offering a high-capacity, secure, and efficient embedding solution, showing strong potential for applications in secure image sharing and cloud storage environments.