A Novel High-Fidelity Reversible Data Hiding Method Based on Adaptive Multi-pass Embedding

Xiaoxi Kong, Wenguang He, Zhanchuan CaiMathematics 2025 · DOI

TL;DR

This paper introduces a novel reversible data hiding (RDH) method that significantly improves image fidelity and embedding efficiency by adaptively leveraging pixel redundancy. By dynamically adjusting prediction parameters based on image content and employing a pairwise prediction error expansion strategy, the method achieves higher PSNR and better distortion control compared to existing techniques.

What Problem Does This Paper Address?

Reversible data hiding (RDH) is crucial for embedding secret information into digital media without altering the original content, enabling perfect restoration. A key challenge in RDH is generating accurate predictions of pixel values to create prediction errors. These errors are then modified to embed data. While methods like Pixel Value Ordering (PVO) offer high fidelity, they often fail to fully exploit the inherent redundancy in image pixels, leading to suboptimal embedding efficiency and increased distortion. Existing PVO-based methods often use fixed parameters or simple adaptive strategies, which may not be optimal for diverse image content. This paper addresses the need for a more sophisticated adaptive approach that can better exploit pixel correlations and minimize distortion, especially for high embedding capacities.

Methods

The proposed method builds upon the Pixel Value Ordering (PVO) framework but introduces several key innovations:

  • Adaptive Multi-pass Prediction: Instead of fixed prediction parameters, the method dynamically selects parameters based on pixel content. This involves analyzing the continuity and spatial correlation of pixels within a block to choose the most appropriate prediction strategy.
  • Exploiting Consecutive Pixel Redundancy: A novel prediction technique is introduced that specifically leverages the redundancy between consecutive pixels, a feature often overlooked by traditional methods.
  • Pairwise Prediction Error Expansion (PEE): The method incorporates pairwise PEE, where two prediction errors are combined to embed data. This allows for more expandable prediction errors and a reduction in distortion compared to single-error modification.
  • Content-Based Adaptive Strategy: The core of the method is an adaptive strategy that dynamically selects multi-pass parameters. This involves classifying pixel blocks based on their noise level (NL) and assigning an optimal parameter 'k' (denoted as h(v)) to each class. This ensures that embedding is tailored to the complexity of different image regions.
  • Threshold Determination: A framework is presented to determine optimal thresholds (Tk*) for classifying pixel blocks, balancing embedding capacity (EC) with distortion (D(Tk)).

Datasets

The paper evaluates its method on standard test images:

  • USC-SIPI database: Includes images like Airplane, Baboon, Barbara, Boat, Elaine, Lake, Lena, and Peppers.
  • Kodak dataset: Used for further validation.

No new benchmarks were introduced; the evaluation focuses on comparing performance against existing methods on these standard datasets.

Key Insights and Ideas

  • Redundancy is Key: The paper strongly emphasizes that exploiting the redundancy between consecutive and similar pixels is crucial for improving both embedding efficiency and image quality.
  • Adaptivity Pays Off: Dynamically adjusting prediction parameters based on local image content (smooth vs. textured regions) leads to significantly better results than fixed-parameter approaches.
  • Pairwise PEE for Efficiency: Combining prediction errors in pairs (pairwise PEE) offers a more efficient way to embed data and reduce distortion, especially for higher capacities.
  • Content-Aware Classification: Classifying pixel blocks by their noise level and assigning specific embedding parameters to each class is a robust way to manage distortion and maximize capacity.

Key Results

  • Improved Fidelity: The proposed method, particularly the PairwiseKP-IPVO variant, consistently achieves higher Peak Signal-to-Noise Ratio (PSNR) compared to state-of-the-art methods across various test images and embedding capacities. For example, on Lena with a 10,000-bit payload, the proposed method achieved 61.56 dB, outperforming others.
  • Competitive Embedding Capacity: While prioritizing fidelity, the method maintains competitive embedding capacities.
  • Reduced Distortion: Experimental results show substantial Mean Squared Error (MSE) reductions, indicating less distortion in the marked images. For instance, compared to PVO and IPVO, MSE was reduced significantly.
  • Performance on Smooth Images: The method shows a particular advantage on smooth images due to its ability to better exploit pixel continuity.

Limitations and Future Work

  • Computational Complexity: The adaptive parameter selection and multi-pass embedding strategy contribute to higher computational complexity compared to simpler methods.
  • Future Work: The authors plan to focus on developing more lightweight iterations of the adaptive parameter selection mechanism. They also intend to explore the potential of high-dimensional histogram modification for RDH.

Summary updated

Abstract

In reversible data hiding, prediction error generation plays a crucial role, with pixel value ordering (PVO) standing out as a prediction method that achieves high fidelity. However, conventional PVO approaches select predicted pixels and their predictions independently, failing to fully exploit the inherent redundancy in ordered pixel sequences. This paper proposes a novel PVO-based prediction method that leverages the continuity and spatial correlation of ordering pixels. We first introduce a new prediction technique that exploits the redundancy of consecutive pixels. Our approach selects the most appropriate prediction method from preset prediction errors, considering both pixel position and value characteristics. Furthermore, we implement an adaptive strategy that dynamically selects multiple iteration parameters based on pixel content to obtain more expandable prediction errors and adjusts the modification of prediction errors accordingly. Unlike traditional fixed-parameter methods, our approach better utilizes the inherent structure and redundancy of image pixels, thereby improving data embedding efficiency while minimizing image distortion. We enhance performance by combining pairwise prediction-error expansion with content-based prediction error analysis. Experimental results demonstrate that the proposed scheme outperforms state-of-the-art solutions in terms of image fidelity while maintaining competitive embedding capacity, confirming the effectiveness of our method for efficient data embedding and image recovery.
BibTeX
@inproceedings{kong2025novel,
  author = {Xiaoxi Kong and Wenguang He and Zhanchuan Cai},
  title = {{A Novel High-Fidelity Reversible Data Hiding Method Based on Adaptive Multi-pass Embedding}},
  booktitle = {Mathematics},
  year = {2025},
  doi = {10.3390/math13111881},
}
Figure