Reversible data hiding based on reducing invalid shifting of pixels in histogram shifting

Yujie Jia, Zhaoxia Yin, Xinpeng Zhang, Yonglong Luo

Abstract

Reversible Data Hiding (RDH) is a specialized steganographic technique where a secret message is embedded into a digital image such that the original "cover" image can be perfectly reconstructed after extraction. This is vital for sensitive applications like medical imaging and forensics. This paper presents a refined RDH approach that minimizes image distortion by prioritizing "smooth" regions for data embedding. By reducing "invalid shifting pixels" (ISPs) during the histogram shifting process, the proposed method achieves higher data capacity and superior visual quality compared to state-of-the-art techniques.

Introduction

The primary goal of Reversible Data Hiding (RDH) is to embed information into a carrier medium without causing permanent alterations. Most modern RDH methods rely on Histogram Shifting (HS), which identifies peaks in a prediction error histogram and shifts values to create space for secret bits. However, standard HS methods often ignore local image characteristics, leading to inefficiencies in textured regions.

The Problem: Invalid Shifting Pixels (ISPs)

Traditional HS methods typically process pixels in a fixed scanning order. When data is embedded in "unsmooth" or highly textured regions (e.g., hair or grass), many pixels are modified or shifted to make room for data but do not actually end up carrying any secret bits. These are termed Invalid Shifting Pixels (ISPs).

Conceptual diagram of histogram shifting highlighting effective vs. invalid pixel shifting.

As shown in the figure above, invalid shifting contributes to unnecessary noise and distortion, reducing the Peak Signal-to-Noise Ratio (PSNR) of the stego-image. The authors demonstrate that embedding data in smooth regions creates far fewer ISPs than embedding in textured regions, confirming that texture-awareness is the key to reducing distortion.

Proposed Methodology

The authors introduce a multi-step pipeline designed to optimize the embedding process by targeting smooth regions first.

1. Checkerboard Division and Adaptive Prediction

The image is first divided into two sets (A and B) using a checkerboard pattern. This spatial separation allows pixels in one set to be predicted using the context provided by the other.

Checkerboard division of image pixels for adaptive prediction.

Instead of simple averaging, a weighted prediction based on four immediate neighbors is used to estimate a pixel's value. The difference between the actual and estimated value forms the prediction error. By using four neighbors, the authors create a much "steeper" prediction error histogram, which directly translates to higher embedding capacity.

2. Texture-Aware Sorting

For every pixel, a "fluctuation value" is calculated to measure local complexity. Pixels are then sorted by these values. The core innovation lies in Sorted Embedding: data is tucked into pixels with the lowest fluctuations first. This "smooth-first" strategy ensures that the most stable regions of the image are utilized, drastically reducing the occurrence of ISPs.

3. Double Peak Strategy

To further enhance capacity, the method identifies two peaks in the prediction error histogram. This effectively doubles the available "slots" for data compared to traditional single-peak methods.

Key Results

The method was evaluated using standard benchmark images like Lena, Baboon, and Airplane. Key findings include:

  • Superior Visual Quality: For the Airplane image, the proposed method outperformed competitors by approximately 5 dB at low payloads.
  • Increased Capacity: On the Lena image, the embedding capacity reached approximately 74,000 bits, nearly double that of some existing methods.
  • Efficiency in ISP Reduction: In a test with a 10,000-bit payload on the Airplane image, the proposed method produced only 3,220 ISPs, compared to over 17,500 ISPs in competing schemes.

Conclusion

By integrating texture-awareness into the histogram shifting process, this research demonstrates that prioritizing smooth regions can significantly mitigate image distortion. The reduction of invalid shifting pixels allows for a more efficient use of the image's spatial redundancy, providing a robust solution for high-capacity, high-fidelity reversible data hiding.

Summary updated

BibTeX
@article{jia2019reversible,
  author = {Yujie Jia and Zhaoxia Yin and Xinpeng Zhang and Yonglong Luo},
  title = {{Reversible data hiding based on reducing invalid shifting of pixels in histogram shifting}},
  journal = {Signal Processing},
  year = {2019},
  doi = {10.1016/j.sigpro.2019.05.020},
}
Figure