TL;DR
This paper introduces a novel Reversible Data Hiding in Encrypted Images (RDH-EI) scheme called CFSS-RDHEI. It leverages a new Cipher-Feedback Secret Sharing (CFSS) technique, inspired by AES, to encrypt images into multiple smaller shares. This allows multiple data hiders to embed secret information independently, with the original image being recoverable from a subset of these shares, offering enhanced security and efficiency.
What Problem Does This Paper Address?
In today's cloud-centric world, protecting sensitive image data while enabling operations like data hiding is crucial. Reversible Data Hiding in Encrypted Images (RDH-EI) aims to achieve this by allowing secret data to be embedded into an encrypted image, with the ability to perfectly restore both the original image and the embedded data. Existing RDH-EI schemes, particularly those using secret sharing for security, often suffer from either lightweight security levels that are vulnerable to attacks or significant data expansion, making them impractical. This paper addresses the need for an RDH-EI scheme that offers robust security, high embedding capacity, and reduced data expansion, especially in scenarios involving multiple data hiders.
Methods
The core innovation is the Cipher-Feedback Secret Sharing (CFSS) technique. Unlike traditional polynomial-based secret sharing, CFSS incorporates a cipher-feedback strategy similar to AES. This means that the output of one sharing operation influences the next, creating a diffusion effect. This diffusion property significantly enhances security by making the shares highly sensitive to changes in the original data, thus resisting attacks like differential and chosen-plaintext attacks.
Building upon CFSS, the authors propose the CFSS-RDHEI scheme. Here's how it works:
- Image Encryption and Sharing: The original image is encrypted using the CFSS technique, generating n encrypted image shares. Each share is significantly smaller than the original image (specifically, 1/(r-1) the size, where r is the threshold).
- Side Information Generation: To enable perfect reconstruction, crucial side information (like prediction errors and location maps for pixel prediction) is generated. This side information is also encrypted using CFSS.
- Data Hiding: Each data hider receives one encrypted image share and its corresponding side information share. They can then embed their secret data into the encrypted image without needing the original image or the encryption key. A multi-MSB prediction method with a secure location map is used for embedding to avoid security issues found in prior methods.
- Image Reconstruction: The receiver needs at least r marked encrypted images and the encryption key. They first reconstruct the side information using CFSS, then use it to recover the original image's multi-MSBs and (8-1)-LSBs, ultimately reconstructing the original image losslessly.
Datasets
The paper evaluates its proposed scheme using eight commonly used grayscale images of size 512x512. These include:
- Lena
- Baboon
- Jetplane
- Man
- Airplane
- Peppers
- Goldhill
- Boat
The secret data embedded into these images was randomly generated.
Key Insights and Ideas
- Cipher-Feedback for Enhanced Security: The integration of a cipher-feedback mechanism into secret sharing, inspired by AES, is a key novelty. This approach significantly boosts the security of the encrypted shares, making them resistant to common cryptanalytic attacks that plague simpler secret sharing schemes.
- Reduced Data Expansion: By making each encrypted share smaller than the original image (1/(r-1) size), the scheme achieves a substantial reduction in data expansion, a critical factor for practical applications.
- Multi-Hider Efficiency: The scheme is designed for multiple data hiders, allowing for parallel embedding of secret data into different shares, which can improve overall throughput.
- Robust Pixel Prediction: The use of a multi-MSB prediction method with a secure location map addresses known security vulnerabilities in previous RDH-EI methods that relied on simpler error flags.
Key Results
- High Security: The CFSS-RDHEI demonstrates high security by achieving NPCR (Number of Pixel Change Rate) and UACI (Uniform Average Change Intensity) values close to theoretical optima, indicating strong resistance to differential attacks. SRCC (Spearman's Rank Correlation Coefficient) and KRCC (Kendall Rank Correlation Coefficient) values close to zero further support this.
- Reduced Data Expansion: The data expansion rate for each data hider is 1/(r-1) of the original image size, which is significantly smaller than other secret sharing-based RDH-EI methods. For example, with r=3, the expansion is 1/2.
- Competitive Embedding Rates: The scheme achieves competitive embedding rates, with values ranging from 1.25 bpp to 3.64 bpp for various test images, depending on the image's characteristics and the chosen (r, n) threshold.
- Lossless Reconstruction: Experimental results confirm that the scheme achieves perfect reconstruction of the original image without any loss.
Limitations and Future Work
- Parameter Sensitivity: The embedding rate is shown to be slightly different for different values of r and n, suggesting that optimal parameter selection is important for maximizing embedding capacity.
- Computational Cost: While the paper highlights reduced data expansion, the computational cost of the CFSS and the multi-MSB prediction method is not explicitly detailed in terms of time complexity.
- Future Work: The paper suggests that the CFSS-RDHEI's ability to resist potential attacks can be further improved by exploring more advanced chaotic systems for pseudorandom number generation. The authors also imply that further optimization of the multi-MSB prediction and embedding strategies could lead to even higher embedding rates.
Summary updated