TL;DR
This paper introduces a novel reversible data hiding (RDH) scheme for Absolute Moment Block Truncation Coding (AMBTC) compressed images. By introducing a difference table and employing a multi-layer embedding strategy for bitmaps, the scheme significantly boosts embedding capacity while maintaining perfect image reversibility, outperforming existing state-of-the-art methods.
What Problem Does This Paper Address?
In today's data-driven world, efficient and secure transmission of digital information is paramount. Images, being a significant portion of this data, present two key challenges: reducing transmission bandwidth and ensuring the security of embedded confidential data. Absolute Moment Block Truncation Coding (AMBTC) is a popular image compression technique that offers a good balance between compression ratio and image quality. However, embedding secret data into AMBTC-compressed images (reversible data hiding) to enhance security for purposes like image management or authentication, often leads to a trade-off between embedding capacity and image quality. As the demand for embedding larger amounts of secret data grows, there's a pressing need for RDH schemes that can achieve higher embedding capacities without compromising image quality or the ability to perfectly restore the original image. Existing AMBTC-based RDH schemes have made progress, but there's still room for improvement in maximizing embedding capacity.
Methods
The proposed scheme builds upon the AMBTC compression framework, which divides an image into blocks and represents each block with a bitmap, a high value, and a low value. The core innovations lie in:
- Difference Table: A new "difference table" is introduced, derived from the difference between the high and low values of each AMBTC block. This table, along with the original high and low tables, forms a set of three value tables.
- Optimized Value Table Embedding: The scheme intelligently selects two out of these three value tables for data embedding, based on which combination yields the highest embedding capacity. This selection is indicated by a 2-bit indicator. Data is embedded into the chosen tables using a Joint Neighborhood Coding (JNC) approach, enhanced with arithmetical and logical differential coding.
- Multi-Layer Bitmap Embedding: For the bitmaps, a three-layer embedding strategy is employed. The first two layers utilize a side-match (SM) technique, similar to existing methods, to embed data by predicting bits based on neighboring blocks. The crucial addition is a third layer that applies Huffman coding to compress any remaining unprocessed bits in the bitmap, thereby creating additional space for embedding more secret data.
- Reversibility: The entire process is designed to be reversible. The embedded data can be extracted, and the original AMBTC compressed file can be perfectly reconstructed. This reconstruction involves a two-stage process for value tables and a multi-layer approach for bitmaps, mirroring the embedding process.
Datasets
The paper evaluates the proposed scheme using a variety of images, including:
- Standard grayscale images (e.g., Airplane, Boat, Goldhill, Peppers).
- Complex grayscale images (e.g., Baboon, Flowers, Mountain, Owl).
- Color images (e.g., Airplane, Baboon, Lake, Peppers).
The experiments were conducted on images of different sizes: 256x256, 512x512, and 1024x1024 pixels. No new benchmark datasets were introduced.
Key Insights and Ideas
- The Power of Differences: Introducing a difference table derived from the high and low values significantly expands the embedding capacity of the value tables. This highlights that exploiting inter-dependencies within the compressed data can unlock more embedding potential.
- Strategic Table Selection: Dynamically choosing the best combination of value tables (high, low, difference) for embedding, rather than relying on a fixed set, is a clever way to maximize embedding capacity on a per-image or per-block basis.
- Huffman for Bitmap Space: Using Huffman coding as a third layer for bitmap embedding is an effective strategy to reclaim space that would otherwise be unused, further boosting the overall embedding rate. This shows that combining different compression and embedding techniques can yield synergistic benefits.
- Trade-off Management: The scheme demonstrates that it's possible to achieve significantly higher embedding rates (over 30%) while maintaining perfect reversibility, suggesting that the perceived trade-off between capacity and reversibility might be less rigid than previously thought with optimized techniques.
Key Results
The proposed scheme demonstrates superior performance compared to state-of-the-art methods:
- Embedding Rate (ER): The scheme consistently achieves higher ERs across various image types and sizes. For standard grayscale images, ERs exceed 30%, with one instance reaching 32.87%. For color images, the ER for "Airplane" reaches 32.87%. For the 1024x1024 "Airplane" image, the ER reaches an impressive 40.00%.
- Pure Embedding Capacity (PEC): The PEC is also significantly improved, indicating a larger amount of secret data can be embedded relative to the original image size. For the 1024x1024 "Airplane" image, the PEC reaches 491,892 bits.
- Performance Comparison: Compared to Liu et al.'s [23] scheme (published in 2024), the proposed scheme shows an ER approximately 2% higher and a PEC about 15,000 bits greater.
- Image Quality: The PSNR values remain identical to the original AMBTC compressed images, confirming perfect reversibility and no degradation in visual quality.
Limitations and Future Work
- Time Complexity for Bitmaps: The paper acknowledges that the bitmap embedding and reconstruction stages have a higher time complexity of O(n x 7 x 7!) due to the need to enumerate permutations. This is presented as a trade-off for achieving higher embedding capacity.
- Future Work: The authors suggest further exploration of the carrier characteristics of AMBTC and investigating schemes for further improvements. This implies that there might be more avenues to explore within the AMBTC framework or by combining it with other techniques to push the boundaries of RDH even further.
Summary updated