Scholarlink

In today’s digital world, data security and integrity are important. Steganography, the process of hiding secret information within seemingly secure media, has gained a lot of attention as a method of obscure communication. Traditional approaches have relied on minor alterations to images, but the recent developments in deep learning led to a new era of innovation in this field. The study focuses on Convolutional Neural Networks with Autoencoders and Decoders (CNN-AD). These CNN-based models embed data into images while maintaining visual coherence, avoiding detection by both steganalytic algorithms and humans. We analyze key features of deep steganography, such as computing efficiency, resilience to attacks, and data concealment capabilities. Our technique revolves around an autoencoder architecture which consists of an encoder and multi decoders, trained on ImageNet dataset. Experimental findings show that our technique is effective, with a high Peak Signal to Noise Ratio (PSNR) of 38.55 and a Structural Similarity Index (SSIM) of 0.985 for the stego image and 37.01, 0.978 for the extracted image, respectively. These results highlight the potential of merging deep learning and large-scale datasets to enhance the field of image steganography, offering new paths for secure and efficient communication.

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

CNN-Based Autoencoder for Image Steganography: A Deep Learning Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • G. Kalaiarasi,
  • A. S. Narmadha,
  • G. Mahalakshmi,
  • Vijai Meyyappan,
  • B. Saritha,
  • B. Rubini

摘要

In today’s digital world, data security and integrity are important. Steganography, the process of hiding secret information within seemingly secure media, has gained a lot of attention as a method of obscure communication. Traditional approaches have relied on minor alterations to images, but the recent developments in deep learning led to a new era of innovation in this field. The study focuses on Convolutional Neural Networks with Autoencoders and Decoders (CNN-AD). These CNN-based models embed data into images while maintaining visual coherence, avoiding detection by both steganalytic algorithms and humans. We analyze key features of deep steganography, such as computing efficiency, resilience to attacks, and data concealment capabilities. Our technique revolves around an autoencoder architecture which consists of an encoder and multi decoders, trained on ImageNet dataset. Experimental findings show that our technique is effective, with a high Peak Signal to Noise Ratio (PSNR) of 38.55 and a Structural Similarity Index (SSIM) of 0.985 for the stego image and 37.01, 0.978 for the extracted image, respectively. These results highlight the potential of merging deep learning and large-scale datasets to enhance the field of image steganography, offering new paths for secure and efficient communication.