[Paper Review] An Improved Reversible Data Hiding in Encrypted Images using Parametric Binary Tree Labeling

Youqing Wu, Youzhi Xiang+3 more|arXiv (Cornell University)|May 23, 2019

Advanced Steganography and Watermarking Techniques25 references134 citations

TL;DR

This paper proposes IPBTL-RDHEI, a high-capacity reversible data hiding scheme in encrypted images using parametric binary tree labeling. By reserving embedding room in the entire original image before encryption and leveraging spatial correlation across the full image (not just small blocks), the method labels encrypted pixels into two categories via a parametric binary tree, enabling bit-replacement-based secret data embedding. The approach achieves higher embedding rates than state-of-the-art methods while ensuring lossless recovery of both the original image and embedded data.

ABSTRACT

This work proposes an improved reversible data hiding scheme in encrypted images using parametric binary tree labeling(IPBTL-RDHEI), which takes advantage of the spatial correlation in the entire original image but not in small image blocks to reserve room for hiding data. Then the original image is encrypted with an encryption key and the parametric binary tree is used to label encrypted pixels into two different categories. Finally, one of the two categories of encrypted pixels can embed secret information by bit replacement. According to the experimental results, compared with several state-of-the-art methods, the proposed IPBTL-RDHEI method achieves higher embedding rate and outperforms the competitors. Due to the reversibility of IPBTL-RDHEI, the original plaintext image and the secret information can be restored and extracted losslessly and separately.

Motivation & Objective

• To address the limited embedding capacity in existing reversible data hiding in encrypted images (RDHEI) schemes that rely on local redundancy in small image blocks.
• To improve embedding rate by exploiting spatial correlation across the entire original image rather than localized blocks.
• To develop a separable and error-free RDHEI scheme where image recovery and secret data extraction can be performed independently.
• To enhance security by ensuring encrypted images reveal no perceptible information, preserving privacy in cloud storage.

Proposed method

• Reserve embedding room in the original plaintext image before encryption using spatial correlation across the entire image.
• Apply a parametric binary tree labeling scheme with parameters α and β to categorize encrypted pixels into two groups: G1 (embeddable) and G2 (non-embeddable).
• Use bit replacement to embed secret data into pixels in the G1 category, where the number of available bits depends on α and β.
• Leverage the full-image spatial correlation to minimize auxiliary information overhead, reducing the need for block-level redundancy.
• Ensure reversibility by storing only minimal auxiliary information (e.g., labeling parameters and bit positions), enabling exact recovery of both original image and secret data.
• Support separable recovery: receivers with different permissions can extract only the secret data, only the original image, or both.

Experimental results

Research questions

• RQ1 Can full-image spatial correlation be leveraged more effectively than block-based correlation to increase embedding capacity in encrypted images?
• RQ2 Can a parametric binary tree labeling scheme enable higher and more flexible embedding rates in RDHEI while maintaining separability and reversibility?
• RQ3 How does the proposed method compare in embedding rate and image quality to state-of-the-art RDHEI schemes, especially in terms of PSNR and SSIM of encrypted images?
• RQ4 To what extent does the method reduce auxiliary information overhead by avoiding block-wise processing?

Key findings

• The proposed IPBTL-RDHEI method achieves a maximum embedding rate of 2.9883 bpp on the BOSSbase and BOWS-2 datasets when α=5 and β=2, significantly outperforming prior methods.
• The average embedding rate across three datasets reaches 2.5613 bpp (BOSSbase), 2.5194 bpp (BOWS-2), and 2.2683 bpp (UCID), demonstrating consistent high performance.
• All reconstructed original images achieve PSNR values approaching +∞ dB and SSIM of 1.0, confirming error-free image recovery.
• Encrypted images maintain extremely low PSNR (around 6.88–9.52 dB) and SSIM (around 0.03–0.07), indicating no perceptible information leakage and strong privacy protection.
• The method outperforms Yi et al.’s [18] block-based approach in both average and maximum embedding rates, especially due to full-image correlation utilization and reduced auxiliary data.
• The scheme is fully separable: secret data and original image can be extracted and restored independently and losslessly.