Noise Removal in Embedded Image With Bit Approximation

Xianquan Zhang, Xuelong Li, Zhenjun Tang, Shichao Zhang, Shaomin Xie

Introduction

In digital steganography, "stego-images" serve as carriers for hidden data, often secret images. However, transmission over noisy channels frequently introduces interference, resulting in corrupted extractions that resemble a "mess of static." Traditional denoising techniques typically employ filters that "blur" away noise by replacing entire pixels. A recent study in IEEE Transactions on Knowledge and Data Engineering proposes a more surgical alternative: rather than replacing the whole pixel, the algorithm identifies and repairs only the specific bits that are broken.

The Challenge of Impulse Noise

When an image is embedded and transmitted, it is susceptible to "impulse noise"—random black and white pixels (salt-and-pepper noise) that degrade visual quality. Standard denoising methods, such as median filters, treat pixels as monolithic units. If a pixel is flagged as noisy, it is discarded and replaced with a value derived from its neighbors.

A corrupted version of the Lena benchmark image exhibiting significant impulse noise.

The authors argue that this approach is inefficient. In a noisy stego-image, a "corrupted" pixel may still contain several accurate bits. Discarding the entire pixel loses valuable original information. The core challenge lies in distinguishing reliable bits from unreliable ones and accurately approximating the latter without compromising the image's texture.

Methodology: Bit-Level Granularity

The proposed $k$NN-bit approximation algorithm shifts the unit of repair from the pixel to the individual bit. The process follows four stages:

  1. Identification: Noise detection on the carrier image flags likely corrupted ("unreliable") bits in the extracted secret image.
  2. MSB Correction: The Most Significant Bit (MSB) is prioritized due to its impact on brightness. The algorithm uses a $k$-nearest neighbors ($k$NN) window, considering only neighbors with reliable high bits to estimate the target's 8th bit.
  3. Mid-Bit Verification (5th–7th Bits): These bits are validated by ensuring the resulting pixel value remains logically bounded between its nearest reliable neighbors, correcting "outliers."
  4. Lower-Bit Approximation: For the remaining bits, the algorithm performs an exhaustive search to find the bit combination that minimizes the difference from the local target value.

Experimental Results

The researchers evaluated the algorithm using standard benchmarks from the USC-SIPI Image Database, including the classic "Lena" image and various aerial and object photos.

The Lena image after restoration using the proposed bit-approximation method.

The bit-approximation method consistently outperformed standard filters. At a high noise density of 0.5, the proposed method achieved a Peak Signal-to-Noise Ratio (PSNR) of 27.12 dB, significantly higher than the 19.91 dB achieved by the next best-compared method. Visually, the algorithm preserved clear edges and textures where traditional filters produced heavy blurring. Furthermore, the restoration ratio for the critical 8th bit remained above 99% across all tested noise levels.

Conclusion and Future Work

While the $k$NN-bit approximation offers superior restoration quality, its primary limitation is computational complexity. The iterative loops and exhaustive searches required for bit-level repair make it slower than simple filtering methods. Future research will focus on optimizing the algorithm for real-time applications—potentially through C++ or GPU acceleration—and extending the logic to handle other forms of degradation like Gaussian blur or compression artifacts.

Summary updated

Abstract

Stego-images are often contaminated by interchannel noise or active noise attack when communicating on the Web. And it is challenging to restore embedded image from corrupted stego-image. This paper studies a <i>k</i> NN-bit approximation algorithm to remove noises in embedded image. The proposed algorithm distinguishes reliable bits from extracted bits, and estimates pixel values by keeping reliable bits unchanged and correcting unreliable bits. Specifically, the 8th (highest) unreliable bit of a pixel can be approximated with its nearest neighbor pixels. And then, if an unreliable bit locates at any one of the <inline-formula><tex-math notation="LaTeX">$5^{th}\sim 7^{th}$</tex-math></inline-formula> bits of a pixel, it is adjusted with two nearest neighbors of the pixel, where the pixel is in-between these two nearest neighbors. Finally, for other unreliable bits, each one is approximated by the maximum and minimum possible values of nearest neighbors of its pixel. We conduct experiments for illustrating the efficiency, and demonstrate that the proposed algorithm can recover the embedded images with good visual quality from corrupted stego-images.
BibTeX
@article{zhang2022noise,
  author = {Xianquan Zhang and Xuelong Li and Zhenjun Tang and Shichao Zhang and Shaomin Xie},
  title = {{Noise Removal in Embedded Image With Bit Approximation}},
  journal = {IEEE Trans. Knowl. Data Eng.},
  year = {2022},
  doi = {10.1109/tkde.2020.2992572},
}
Figure