Fidelity Preserved Data Hiding in Encrypted Highly Autocorrelated Data Based on Homomorphism and Compressive Sensing

Yushu Zhang (张玉书), Ming Li, Lanlan Wang, Jingjing Fan, Yushu Zhang, Kanglei Zhou, Haiju Fan

Introduction

In the era of cloud computing, privacy protection has become a paramount concern for users uploading multimedia content to remote servers. To safeguard sensitive information, images and videos are typically encrypted before transmission. However, cloud service providers often require the ability to embed auxiliary metadata—such as tags, timestamps, or authentication data—directly into these encrypted files for management and verification purposes.

The primary challenge in this domain is that standard encryption algorithms transform structured image data into random-looking noise, effectively destroying the spatial correlation (the tendency for neighboring pixels to share similar values). Without this correlation, finding sufficient "room" to hide extra data without permanently degrading the image quality is difficult. Existing methods frequently result in decrypted images with poor visual fidelity or offer extremely limited embedding capacity.

Proposed Methodology

This paper introduces a multi-stage pipeline designed to balance high security with significant data-hiding capacity. The core innovation lies in the use of homomorphic encryption and compressive sensing to preserve image fidelity while allowing third-party data embedding.

System architecture of the proposed data hiding scheme showing the interaction between the image sender, data hider, and receiver.

1. Homomorphic Encryption and Block Classification

Instead of utilizing standard ciphers, the system employs an additive homomorphic cryptosystem based on modular addition. This allows the data hider to perform texture analysis (calculating smoothness versus roughness) on image blocks directly in the encrypted domain. Because the homomorphic property ensures that calculations in the encrypted domain correspond to those in the plaintext domain, the system can accurately identify "smooth" blocks (low variance) and "textured" blocks (high variance) without decrypting the content.

2. Compressive Sensing (CS)

To create space for additional data, the authors apply Compressive Sensing to the Least Significant Bit (LSB) layers of the identified smooth blocks. CS is a signal processing technique that allows a sparse signal to be reconstructed from fewer samples than traditionally required. By compressing these LSBs, the system vacates space for the additional metadata while maintaining the ability to reconstruct the original pixel values later.

3. Security and Separability

To ensure cryptographic robustness, encryption and permutation keys are generated using a Cascade Chaotic System (CCS), which provides high unpredictability and sensitivity to initial conditions. Furthermore, the system is designed with separability: the encryption key and the data-hiding key are independent. A user possessing only the encryption key can still decrypt a high-quality version of the image, even if they cannot extract the hidden metadata.

Experimental Evaluation

The researchers evaluated the proposed method using a suite of standard grayscale test images frequently used in image processing benchmarks, including Lena, Peppers, Man, Airplane, Baboon, Lake, Camera, and Milkdrop. Tests were conducted across varied resolutions, ranging from $128 \times 128$ to $1024 \times 1024$ pixels, to ensure the scalability of the approach.

Results and Discussion

Visual Quality and Fidelity

The "directly decrypted" images—those decrypted without first removing the hidden data—achieved a Peak Signal-to-Noise Ratio (PSNR) of over 40 dB and a Structural Similarity Index (SSIM) close to 1. These metrics indicate that the hidden data is virtually invisible to the naked eye, preserving the visual truthfulness of the host image.

Security Performance

The encryption showed excellent resistance to statistical attacks. The Number of Pixels Change Rate (NPCR) was approximately 99.6%, and the Information Entropy remained near the ideal value of 8. These results suggest that the encrypted images are indistinguishable from random noise, providing high-level privacy protection.

Comparative Superiority

When compared to several recent state-of-the-art schemes, the proposed method demonstrated a higher embedding capacity and superior decrypted image quality. By targeting smooth areas that are more compressible and less sensitive to change, the system maximizes the efficiency of the data-hiding process.

Conclusion

This research presents a robust framework for fidelity-preserved data hiding in encrypted domains. By leveraging the mathematical advantages of homomorphic encryption and the efficiency of compressive sensing, the authors provide a solution that meets the needs of both privacy-conscious users and cloud service providers. While the current embedding capacity is limited by the inherent sparsity of LSB layers, future work will explore more sophisticated sparsity bases to further enhance reconstruction quality and data capacity.

Summary updated

BibTeX
@article{2019fidelity,
  author = {Yushu Zhang (张玉书) and Ming Li and Lanlan Wang and Jingjing Fan and Yushu Zhang and Kanglei Zhou and Haiju Fan},
  title = {{Fidelity Preserved Data Hiding in Encrypted Highly Autocorrelated Data Based on Homomorphism and Compressive Sensing}},
  journal = {IEEE Access},
  year = {2019},
  doi = {10.1109/access.2019.2919376},
}
Figure