Scholarlink

<p>This paper presents a Reinforcement Learning -based multi-tier bit-plane (RL-MTBP) embedding approach, in which an intelligent agent adaptively selects optimal color channels and tiers for data hiding. Conventional steganographic techniques suffer from limited payload capacity, weak resistance to various attacks, and increased vulnerability to statistical steganalysis. Moreover, they lack adaptive embedding mechanisms and are generally confined to single-channel, single-receiver frameworks. The proposed technique dynamically utilises multiple tiers to improve embedding performance. The RL agent is trained using a reward function based on image quality metrics such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and mean squared error (MSE), ensuring a balance between embedding capacity and imperceptibility. Additionally, secure data embedding is achieved through key-based random pixel selection, enabling independent and robust multi-receiver data hiding while maintaining resistance to steganalysis. RL-MTBP achieves 5 bpp embedding capacity for grayscale images with PSNR in the range of 35–41 dB, and 10 bpp for RGB images with higher visual quality, maintaining PSNR between 43 and 48 dB for 512 × 512 images. The RL-MTBP approach demonstrates strong robustness against various attacks, including Salt &amp; Pepper noise, Impulse noise, Random Pixel Corruption, and Bit Flip attacks. Additionally, it exhibits enhanced security and stability when evaluated using steganalysis techniques such as histogram analysis, Sample Pair Analysis (SPA), and Regular–Singular (RS) analysis. Overall, the RL-MTBP algorithm achieves a well-balanced performance by maintaining acceptable image quality metrics while ensuring high embedding capacity, robustness, and strong security.</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

Reinforcement learning-based adaptive multi-tier bit-plane embedding for high-capacity, robust, and secure image steganography

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月24日

查看全文
  • Gauri Joshi,
  • Shilpa Metkar,
  • Vibha Vyas

摘要

This paper presents a Reinforcement Learning -based multi-tier bit-plane (RL-MTBP) embedding approach, in which an intelligent agent adaptively selects optimal color channels and tiers for data hiding. Conventional steganographic techniques suffer from limited payload capacity, weak resistance to various attacks, and increased vulnerability to statistical steganalysis. Moreover, they lack adaptive embedding mechanisms and are generally confined to single-channel, single-receiver frameworks. The proposed technique dynamically utilises multiple tiers to improve embedding performance. The RL agent is trained using a reward function based on image quality metrics such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and mean squared error (MSE), ensuring a balance between embedding capacity and imperceptibility. Additionally, secure data embedding is achieved through key-based random pixel selection, enabling independent and robust multi-receiver data hiding while maintaining resistance to steganalysis. RL-MTBP achieves 5 bpp embedding capacity for grayscale images with PSNR in the range of 35–41 dB, and 10 bpp for RGB images with higher visual quality, maintaining PSNR between 43 and 48 dB for 512 × 512 images. The RL-MTBP approach demonstrates strong robustness against various attacks, including Salt & Pepper noise, Impulse noise, Random Pixel Corruption, and Bit Flip attacks. Additionally, it exhibits enhanced security and stability when evaluated using steganalysis techniques such as histogram analysis, Sample Pair Analysis (SPA), and Regular–Singular (RS) analysis. Overall, the RL-MTBP algorithm achieves a well-balanced performance by maintaining acceptable image quality metrics while ensuring high embedding capacity, robustness, and strong security.